Assessment of four statistical pattern recognition techniques to assist in obstructive sleep apnoea diagnosis from nocturnal oximetry

The aim of this study is to assess the utility of traditional statistical pattern recognition techniques to help in obstructive sleep apnoea (OSA) diagnosis. Classifiers based on quadratic (QDA) and linear (LDA) discriminant analysis, K-nearest neighbours (KNN) and logistic regression (LR) were evaluated. Spectral and nonlinear input features from oxygen saturation (SaO₂) signals were applied. A total of 187 recordings from patients suspected of suffering from OSA were available. This initial dataset was divided into training set (74 subjects) and test set (113 subjects). Twelve classification algorithms were developed by applying QDA, LDA, KNN and LR with spectral features, nonlinear features and combination of both groups. The performance of each algorithm was measured on the test set by means of classification accuracy and receiver operating characteristic (ROC) analysis. QDA, LDA and LR showed better classification capability than KNN. The classifier based on LDA with spectral features provided the best diagnostic ability with an accuracy of 87.61% (91.05% sensitivity and 82.61% specificity) and an area under the ROC curve (AROC) of 0.925. The proposed statistical pattern recognition techniques could be applied as an OSA screening tool.     

Algorithm for the classification of multi-modulating signals on the electrocardiogram

This article discusses the algorithm to measure electrocardiogram (ECG) and respiration simultaneously and to have the diagnostic potentiality for sleep apnoea from ECG recordings. The algorithm is composed by the combination with the three particular scale transform of a(j)(t), u(j)(t), o(j)(a(j)) and the statistical Fourier transform (SFT). Time and magnitude scale transforms of a(j)(t), u(j)(t) change the source into the periodic signal and tau(j) = o(j)(a(j)) confines its harmonics into a few instantaneous components at tau(j) being a common instant on two scales between t and tau(j). As a result, the multi-modulating source is decomposed by the SFT and is reconstructed into ECG, respiration and the other signals by inverse transform. The algorithm is expected to get the partial ventilation and the heart rate variability from scale transforms among a(j)(t), a(j+1)(t) and u(j+1)(t) joining with each modulation. The algorithm has a high potentiality of the clinical checkup for the diagnosis of sleep apnoea from ECG recordings.     

Night-time sleep and daytime sleepiness in narcolepsy

This report describes night-time sleep and daytime sleepiness in a large (N=530) sample of patients meeting the International Classification of Sleep Disorders criteria for diagnosis of narcolepsy. Sleep data were obtained from polysomnographic recordings on two consecutive nights. Sleepiness was assessed using the Multiple Sleep Latency Test, the Maintenance of Wakefulness Test and the Epworth Sleepiness Scale. Analysis revealed that sleep was mild to moderately disturbed on both recording nights. A first-night effect was suggested by decreased REM latency and increased percentage REM and slow-wave sleep on the second night. Sleepiness and sleep disturbance varied across patient subgroups created based on patient ethnicity and on the presence/absence of cataplexy, sleep apnoea, and periodic limb movements. Covariation of sleep and sleepiness measures across patients was significant but weak. Strong association was found between subgroup means of sleep and sleep disturbance measures. The findings reported here show that sleepiness and sleep disturbance vary across patient subgroups and that sleep disturbance is related to, although unable to account, for the pathological sleepiness of narcolepsy.     

Cross-correlation of EEG frequency bands and heart rate variability for sleep apnoea classification

Sleep apnoea is a sleep breathing disorder which causes changes in cardiac and neuronal activity and discontinuities in sleep pattern when observed via electrocardiogram (ECG) and electroencephalogram (EEG). Using both statistical analysis and Gaussian discriminative modelling approaches, this paper presents a pilot study of assessing the cross-correlation between EEG frequency bands and heart rate variability (HRV) in normal and sleep apnoea clinical patients. For the study we used EEG (delta, theta, alpha, sigma and beta) and HRV (LF_nu, HF_nu and LF/HF) features from the spectral analysis. The statistical analysis in different sleep stages highlighted that in sleep apnoea patients, the EEG delta, sigma and beta bands exhibited a strong correlation with HRV features. Then the correlation between EEG frequency bands and HRV features were examined for sleep apnoea classification using univariate and multivariate Gaussian models (UGs and MGs). The MG outperformed the UG in the classification. When EEG and HRV features were combined and modelled with MG, we achieved 64% correct classification accuracy, which is 2 or 8% improvement with respect to using only EEG or ECG features. When delta and acceleration coefficients of the EEG features were incorporated, then the overall accuracy improved to 71%.     

Design and evaluation of a handheld impedance plethysmograph for measuring heart rate variability

Heart rate variability (HRV) analysis from 10s ECGs has been shown to be reliable. However, the short examination time warrants a user-friendly system that can be used forad-hoc examinations without normal preparation, unlike ECG. A handheld device has been developed that can measure ultra-short HRV from impedance plethysmographic recordings of the pulse wave in distal superficial arteries. The prototype device was made user-friendly through a compact, pen-like design and the use of integrated metal electrodes that were especially designed for dry operation. The main signal processing was performed by a digital signal processor, where the discrete heart beats were detected using a correlation algorithm that could adapt to individual pulse wave shapes to account for biological variation. The novel device was evaluated in 20 mainly young volunteers, using 10 s time-correlated ECG recordings as the reference method. Agreement between the two methods in measuring heart rate and root mean square of successive differences in the heart beat interval (RMSSD) was analysed using correlation coefficients (Pearson's R²), mean differences with 95% confidence intervals and 95% limits of agreement, and Bland-Altman plots. The correlation between the two methods was R²=1.00 and R²=0.99 when heart rate and RMSSD were measured, respectively. The Bland-Altman plots showed suitable agreement between the novel device and standard 10 s ECGs, which was substantiated by 95% limits of agreement of the difference of ± 0.1 beats min⁻¹ and ∼ ± 10 ms for heart rate and RMSSD, respectively. Therefore the evaluation showed no significant systematic error of the novel device compared with ECG.     

Beat-to-beat repolarisation variability in body surface electrocardiograms

The repolarisation variability in body surface electrocardiograms has been evaluated by beat-to-beat QT interval variability. Interpolated R-peak time and template T-wave matching algorithms were used to determine the characteristic time points of the R-wave and T-wave, respectively. The T-wave time can be determined accurately and robustly by searching for the best match between a template T-wave and measured T-waves. The authors studied 5 min multichnnel ECG recordings (35 channels) measured in 20 healthy subjects. A QT variability of 2.24±0.79 ms was obtained (1.15±0.30 ms, if linear detrend was used), which is significantly lower than that reported in several other studies. To explore this discrepancy, the sensitivity of the template matching algorithm to periodic and random noise on the ECG was estimated by a simulation study. The results showed that the repolarisation variability depended on selection of the appropriate lead, the signal-to-noise ratio and the effectiveness of baseline correction. Lead II of a standard 12-lead ECG is a reasonable choice for QT variability analysis; however, precordial leads V3-V6 could be better with regard to the amplitude of the T-wave. Poor signal-to-noise ratios can lead to unrealistic values for repolarisation variability.     

Detection of different recumbent body positions from the electrocardiogram

Changes in body position alter the relative angle between ECG electrodes and the mean electric axis of the heart. These changes influence the time interval during which the projection of the electric dipole, on any ECG lead, is positive (R-wave). In this study, measurements of R-wave duration (RWD) were used to identify changes in body position, and two of its uncorrelated features were used to classify each heartbeat into four basic groups relating to four body positions (supine, prone, left-side, right-side). Data were acquired from healthy volunteers during controlled condition experiments that included well-defined sequences of body positions and simultaneous recordings of ECG leads I, II and III. Results showed over 90% correct identifications ofbody position changes when using any of the three leads. Lead II had the best performance for theclassification of body position and correctly classified 80% of heartbeats. Classification did not improve for a combination of two leads. The technique can be used to reveal additional important clinical information and can be easily implemented, in a variety of applications where ECG is recorded, such as sleep studies, Holter recordings and ischaemia detection.     

Extraction of fetal electrocardiogram using <Emphasis Type="Italic">H</Emphasis><Superscript>∞</Superscript> adaptive algorithms

The fetal electrocardiogram (fECG) contains important information regarding the health of the fetus. However, the fECG obtained noninvasively from the abdominal surface electrical recordings of a pregnant woman are dominated by strong interference from the maternal electrocardiogram (mECG). In this paper, based on the H(infinity) principle, two adaptive algorithms are proposed for the extraction of fECG from the trans-abdominal recordings of pregnant women. The motivation behind the application of H(infinity) techniques is the fact that they are robust with respect to model uncertainties and lack of statistical information regarding noise. The proposed algorithms are applied to simulated as well as real multichannel ECG recordings and their performances are compared to that of the well-known least-mean-square (LMS) adaptive algorithm. It is found that the proposed H(infinity) based algorithms perform superior to the LMS algorithm in extracting the fECG signal.     

Screening for sleep‐disordered breathing with Pediatric Sleep Questionnaire in children with underlying conditions

The Pediatric Sleep Questionnaire described by Chervin et al. (Sleep Medicine, 2000, 1, 21–32) was originally validated for children with obstructive sleep apnoea syndrome but without other disorders. The aim of our study was to check the applicability of this questionnaire in children with underlying chronic medical conditions. Children aged 2–18 years who underwent a diagnostic sleep study at Great Ormond Street Hospital were recruited over a 10‐month period. The Pediatric Sleep Questionnaire completed by their parents and cardiorespiratory polygraphy were scored. Sensitivities and specificities of the Pediatric Sleep Questionnaire were calculated using a Pediatric Sleep Questionnaire score of 0.33 as being indicative of sleep‐disordered breathing. A total of 561 patients were reviewed. Neuromuscular disorders (n = 108), craniofacial anomalies (n = 58) and the obstructive sleep apnea syndrome control group (n = 155) were best represented. The sensitivity for patients with isolated obstructive sleep apnoea syndrome was 76.5% when using an apnoea–hypopnoea index ≥ 5, but this was much lower when looking at specific sub‐groups such as neuromuscular patients (25%) or patients with Trisomy 21 (36.7%). Sensitivities remained unchanged for patients with obstructive sleep apnoea syndrome (77.3%) when an apnoea–hypopnoea index of ≥ 1 was used, but improved for neuromuscular disorders sub‐groups (36.7%) and Trisomy 21 (84%). In conclusion, the Pediatric Sleep Questionnaire is not a good screening tool for obstructive sleep apnoea syndrome in children with complex underlying disorders when a cut‐off apnoea–hypopnoea index of ≥ 5 is used, and it cannot replace cardiorespiratory polygraphy recording.     

一种基于几何特征的ECG波形识别算法

目的 ECG自动分析系统由两部分组成:波形识别和智能诊断。在实际应用中,心电波形识别是该系统的关键。波形识别的精确性和可靠性决定了心脏病诊断的可靠性。为提高波形识别的速率及准确度,本文提出一种基于几何特征的ECG波形识别算法。方法首先利用数字滤波算法对信号进行预处理,提高信号的信噪比,然后通过改进的二阶导数计算出数据的几何特征:点的斜率和运动趋势,并在此基础上,结合ECG波形的实际物理特征,利用算法实现T波、P波、QRS波群的起点、终点以及波峰波谷的自动识别。结果统计分析结果表明,本算法能够快速高效地识别ECG波形。同时将该算法与其他当前各种ECG波形识别算法进行对比,该识别算法在识别的精确性与阳性预测值方面具有更好的性能。结论本文提出的基于几何特征的ECG波形识别算法可以进一步提高当前ECG波形识别算法的性能。     

Automatic analysis of pre‐ejection period during sleep using impedance cardiogram

The pre‐ejection period (PEP) is a valid index of myocardial contractility and beta‐adrenergic sympathetic control of the heart defined as the time between electrical systole (ECG Q wave) to the initial opening of the aortic valve, estimated as the B point on the impedance cardiogram (ICG). B‐point detection accuracy can be severely impacted if ICG cardiac cycles corrupted by motion artifact, noise, or electrode displacement are included in the analyses. Here, we developed new algorithms to detect and exclude corrupted ICG cycles by analyzing their level of activity. PEP was then estimated and analyzed on ensemble‐averaged clean ICG cycles using an automatic algorithm previously developed by the authors for the detection of B point in awake individuals. We investigated the algorithms’ performance relative to expert visual scoring on long‐duration data collected from 20 participants during overnight recordings, where the quality of ICG could be highly affected by movement artifacts and electrode displacements and the signal could also vary according to sleep stage and time of night. The artifact rejection algorithm achieved a high accuracy of 87% in detection of expert‐identified corrupted ICG cycles, including those with normal amplitude as well as out‐of‐range values, and was robust to different types and levels of artifact. Intraclass correlations for concurrent validity of the B‐point detection algorithm in different sleep stages and in‐bed wakefulness exceeded 0.98, indicating excellent agreement with the expert. The algorithms show promise toward sleep applications requiring accurate and reliable automatic measurement of cardiac hemodynamic parameters.     

Automated detection of obstructive sleep apnoea syndrome from oxygen saturation recordings using linear discriminant analysis

Nocturnal polysomnography (PSG) is the gold-standard to diagnose obstructive sleep apnoea syndrome (OSAS). However, it is complex, expensive, and time-consuming. We present an automatic OSAS detection algorithm based on classification of nocturnal oxygen saturation (SaO₂) recordings. The algorithm makes use of spectral and nonlinear analysis for feature extraction, principal component analysis (PCA) for preprocessing and linear discriminant analysis (LDA) for classification. We conducted a study to characterize and prospectively validate our OSAS detection algorithm. The population under study was composed of subjects suspected of suffering from OSAS. A total of 214 SaO₂ signals were available. These signals were randomly divided into a training set (85 signals) and a test set (129 signals) to prospectively validate the proposed method. The OSAS detection algorithm achieved a diagnostic accuracy of 93.02% (97.00% sensitivity and 79.31% specificity) on the test set. It outperformed other alternative implementations that either use spectral and nonlinear features separately or are based on logistic regression (LR). The proposed method could be a useful tool to assist in early OSAS diagnosis, contributing to overcome the difficulties of conventional PSG.     

ECG signal analysis for the assessment of sleep-disordered breathing and sleep pattern

The diagnosis of sleep-disordered breathing (SDB) usually relies on the analysis of complex polysomnographic measurements performed in specialized sleep centers. Automatic signal analysis is a promising approach to reduce the diagnostic effort. This paper addresses SDB and sleep assessment solely based on the analysis of a single-channel ECG recorded overnight by a set of signal analysis modules. The methodology of QRS detection, SDB analysis, calculation of ECG-derived respiration curves, and estimation of a sleep pattern is described in detail. SDB analysis detects specific cyclical variations of the heart rate by correlation analysis of a signal pattern and the heart rate curve. It was tested with 35 SDB-annotated ECGs from the Apnea-ECG Database, and achieved a diagnostic accuracy of 80.5%. To estimate sleep pattern, spectral parameters of the heart rate are used as stage classifiers. The reliability of the algorithm was tested with 18 ECGs extracted from visually scored polysomnographies of the SIESTA database; 57.7% of all 30 s epochs were correctly assigned by the algorithm. Although promising, these results underline the need for further testing in larger patient groups with different underlying diseases.     

Agreement among subjective, objective, and collateral measures of insomnia in postwithdrawal recovering alcoholics

The level of agreement among objective, subjective, and collateral assessments of insomnia was examined in 56 recovering alcoholics. Participants underwent a multimodal sleep assessment protocol consisting of sleep logs, actigraph recordings, questionnaires, and collateral reports of insomnia severity. All sleep measures confirmed moderate to severe insomnia in the study sample. Over 1 week of simultaneous sleep log and actigraph recording, the average disagreement between methods ranged from 16 min for sleep onset latency to 1 hr for wake time after sleep onset. Interrater agreement for the severity of insomnia symptoms using the Sleep Impairment Index was poor for subject-clinician, subject-collateral, and collateral-clinician rating pairs (intraclass correlation coefficients < .35). In general, recovering alcoholics' self-reported sleep reflected a greater severity of insomnia symptoms than did the actigraph and collateral measures. Given that such high levels of disagreement can occur in individual participants, researchers are advised to use a combination of sleep measures to assess insomnia in this population.     

Agreement Among Subjective,Objective, and Collateral Measures of Insomnia in Postwithdrawal Recovering Alcoholics

The level of agreement among objective, subjective, and collateral assessments of insomnia was examined in 56 recovering alcoholics. Participants underwent a multimodal sleep assessment protocol consisting of sleep logs, actigraph recordings, questionnaires, and collateral reports of insomnia severity. All sleep measures confirmed moderate to severe insomnia in the study sample. Over 1 week of simultaneous sleep log and actigraph recording, the average disagreement between methods ranged from 16 min for sleep onset latency to 1 hr for wake time after sleep onset. Interrater agreement for the severity of insomnia symptoms using the Sleep Impairment Index was poor for subject-clinician, subject-collateral, and collateral-clinician rating pairs (intraclass correlation coefficients < .35). In general, recovering alcoholics' self-reported sleep reflected a greater severity of insomnia symptoms than did the actigraph and collateral measures. Given that such high levels of disagreement can occur in individual participants, researchers are advised to use a combination of sleep measures to assess insomnia in this population.     

A comparison of photoplethysmography and ECG recording to analyse heart rate variability in healthy subjects

Measures of heart rate variability (HRV) are widely used to assess autonomic nervous system (ANS) function. The signal from which they are derived requires accurate determination of the interval between successive heartbeats; it can be recorded via electrocardiography (ECG), which is both non-invasive and widely available. However, methodological problems inherent in the recording and analysis of ECG traces have motivated a search for alternatives. Photoplethysmography (PPG) constitutes another means of determining the timing of cardiac cycles via continuous monitoring of changes in blood volume in a portion of the peripheral microvasculature. This technique measures pulse waveforms, which in some instances may prove a practical basis for HRV analysis. We investigated the feasibility of using earlobe PPG to analyse HRV by applying the same analytic process to PPG and ECG recordings made simultaneously. Comparison of 5-minute recordings demonstrated a very high degree of correlation in the temporal and frequency domains and in nonlinear dynamic analyses between HRV measures derived from PPG and ECG. Our results confirm that PPG provides accurate interpulse intervals from which HRV measures can be accurately derived in healthy subjects under ideal conditions, suggesting this technique may prove a practical alternative to ECG for HRV analysis. This finding is of particular relevance to the care of patients suffering from peripheral hyperkinesia or tremor, which make fingertip PPG recording impractical, and following clinical interventions known to introduce electrical artefacts into the electrocardiogram.     

Beyond K-complex binary scoring during sleep: probabilistic classification using deep learning

Study ObjectivesK-complexes (KCs) are a recognized electroencephalography marker of sensory processing and a defining feature of sleep stage 2. KC frequency and morphology may also be reflective of sleep quality, aging, and a range of sleep and sensory processing deficits. However, manual scoring of K-complexes is impractical, time-consuming, and thus costly and currently not well-standardized. Although automated KC detection methods have been developed, performance and uptake remain limited.MethodsThe proposed algorithm is based on a deep neural network and Gaussian process, which gives the input waveform a probability of being a KC ranging from 0% to 100%. The algorithm was trained on half a million synthetic KCs derived from manually scored sleep stage 2 KCs from the Montreal Archive of Sleep Study containing 19 healthy young participants. Algorithm performance was subsequently assessed on 700 independent recordings from the Cleveland Family Study using sleep stages 2 and 3 data.ResultsThe developed algorithm showed an F1 score (a measure of binary classification accuracy) of 0.78 and thus outperforms currently available KC scoring algorithms with F1 = 0.2–0.6. The probabilistic approach also captured expected variability in KC shape and amplitude within individuals and across age groups.ConclusionsAn automated probabilistic KC classification is well suited and effective for systematic KC detection for a more in-depth exploration of potential relationships between KCs during sleep and clinical outcomes such as health impacts and daytime symptomatology.     

Apnea and heart rate detection from tracheal body sounds for the diagnosis of sleep-related breathing disorders

Sleep apnea is one of the most common sleep disorders. Here, patients suffer from multiple breathing pauses longer than 10 s during the night which are referred to as apneas. The standard method for the diagnosis of sleep apnea is the attended cardiorespiratory polysomnography (PSG). However, this method is expensive and the extensive recording equipment can have a significant impact on sleep quality falsifying the results. To overcome these problems, a comfortable and novel system for sleep monitoring based on the recording of tracheal sounds and movement data is developed. For apnea detection, a unique signal processing method utilizing both signals is introduced. Additionally, an algorithm for extracting the heart rate from body sounds is developed. For validation, ten subjects underwent a full-night PSG testing, using the developed sleep monitor in concurrence. Considering polysomnography as gold standard the developed instrumentation reached a sensitivity of 92.8% and a specificity of 99.7% for apnea detection. Heart rate measured with the proposed method was strongly correlated with heart rate derived from conventional ECG (r ² = 0.8164). No significant signal losses are reported during the study. In conclusion, we demonstrate a novel approach to reliably and noninvasively detect both apneas and heart rate during sleep.     

Knowledge-based automatic sleep-stage recognition - reduction in the interpretation variability

Summary Question  The potential of an adaptive algorithm approach for solving existing problems in interpretation variability for sleep-stage recognition and the low acceptance of automatic systems were assessed. Methods  The rules defined by Rechtschaffen and Kales are modelled by a neuro-fuzzy system with the ARTISANA (Artificial Intelligence in Sleep Analysis) algorithm investigated here. The system was trained on the basis of 10 recordings and corresponding hypnograms from a first human scorer. Results  The system agreed with the first scorer at a rate typical for the range for patients with sleep disorders (with a validation set of 28 recordings), with no systematic deviations. Compared with a second scorer, the first scorer and the trained system agreed less. Stage wake was systematically overestimated and deep sleep was systematically underestimated. After adding four additional recordings to the training set, which were classified by the second scorer, the automatic system reached higher agreement rates with both scorers (72.7 % and 73.0 %, respectively) than the scorers did with each other (69.3 %). Compared with the consensus epochs of the two human experts, a very high agreement rate (84.4 %) was found. Conclusion  The results show that the ARTISANA algorithm can reduce systematic variations and produce objective and completely reproducible hypnograms in clinical practice after training with recordings from a set of experienced scorers.     

Disturbed sleep while being on-call: an EEG study of ships' engineers

In order to investigate the effects of on-call duty on sleep and wakefulness, five male ships' engineers were studied using electroencephalogram (EEG) and electrocardiogram (ECG) recordings and subjective ratings. Sleep during on-call nights (two alarms) was shortened and contained less slow wave sleep (SWS) and rapid eye movement (REM) sleep, lower spectral power density, and a higher heart rate. Many of the effects were observable before any alarms had occurred. Rated sleep quality was lower, and sleepiness was higher during the subsequent day. It was suggested that the effects were due to apprehension/uneasiness induced by the prospect of being awakened by an alarm.