Acoustic Estimation of Neck Fluid Volume

Recently we showed that fluid accumulation in the neck can narrow the upper airway (UA) and increase its collapsibility, which may exacerbate obstructive sleep apnea (OSA). However, the available methods for measuring neck fluid volume (NFV) are inconvenient and expensive. Narrowing of the UA due to fluid accumulation could change acoustic characteristics of respiratory sounds. In this study, we developed a novel approach for non-invasive estimation of NFV from acoustic measurements. Twenty-eight healthy subjects lay awake and supine for 90 min while NFV and tracheal sounds were measured simultaneously using bioimpedance and a microphone, respectively. Sets of tracheal sound features were calculated in time and frequency domains and were reduced using methods based on regression and minimum-redundancy-maximum-relevance. The resulting feature sets were applied to a multi-linear regression and a mixture-density neural network to estimate NFV. Our results show very small relative estimation errors of 1.25 and 3.23%, based on the regression and neural network methods, respectively. These results support the practical application of this technology in diagnosing fluid accumulation in the neck and its possible contributions to the pathogenesis of OSA.     

Lower body positive pressure increases upper airway collapsibility in healthy subjects

We previously showed that rostral fluid displacement by lower body positive pressure (LBPP) narrows the upper airway (UA) and increases UA resistance, but effects on UA collapsibility remained unknown. To test if LBPP increases UA collapsibility, 13 healthy men were randomized into a control or LBPP arm then crossed over into the other arm with a 30-min washout in between. LBPP was applied by inflating anti-shock trousers wrapped around both legs to 40 mmHg. UA collapsibility was assessed by determining UA critical closing pressure (P crit) during application of different negative airway pressures. P crit and leg fluid volume were measured at baseline and after 5 min during both periods. LBPP caused a significant increase in P crit associated with a reduction in leg fluid volume. We conclude that rostral fluid displacement by LBPP increases UA collapsibility in healthy men, suggesting that fluid shift into the neck could increase UA collapsibility during sleep and thereby predispose patients with fluid overload states to obstructive sleep apnea.     

Automatic breath and snore sounds classification from tracheal and ambient sounds recordings

In this study respiratory sound signals were recorded from 23 patients suspect of obstructive sleep apnea, who were referred for the full-night sleep lab study. The sounds were recorded with two microphones simultaneously: one placed over trachea and one hung in the air in the vicinity of the patient. During recording the sound signals, patients’ Polysomnography (PSG) data were also recorded simultaneously. An automatic method was developed to classify breath and snore sound segments based on their energy, zero crossing rate and formants of the sound signals. For every sound segment, the number of zero crossings, logarithm of the signal's energy and the first formant were calculated. Fischer Linear Discriminant was implemented to transform the 3-dimensional (3D) feature set to a 1-dimensional (1D) space and the Bayesian threshold was applied on the transformed features to classify the sound segments into either snore or breath classes. Three sets of experiments were implemented to investigate the method's performance for different training and test data sets extracted from different neck positions. The overall accuracy of all experiments for tracheal recordings were found to be more than 90% in classifying breath and snore sounds segments regardless of the neck position. This implies the method's accuracy is insensitive to patient's position; hence, simplifying data analysis for an entire night recording. The classification was also performed on sounds signals recorded simultaneously with an ambient microphone and the results were compared with those of the tracheal recording.     

用于睡眠呼吸暂停检测的心电特征稳定性分析

基于心电信号(electrocardiogram,ECG)的睡眠呼吸暂停检测具有十分重要的意义,很多研究致力于提高检测的准确率却忽视了特征的稳定性。本研究对用于睡眠呼吸暂停检测的心电特征进行稳定性分析,并建立呼吸暂停事件检测模型。基于集成稳定特征选择策略,将最小冗余最大相关(minimal-redundancy-maximal-relevance,mRMR)特征选择方法与稳健排序聚合(robust rank aggregation,RRA)方法结合,对45个心电特征进行稳定性分析。使用10折交叉验证及支持向量机(SVM)进行特征验证及检测模型建立。最终使用14个特征建立分类模型,在独立测试集上实现Acc=90.03%,Se=86.71%,Sp=91.73%,所选特征在稳定性及检测准确率方面有明显提高。     

Feature selection and classifier performance in computer-aided diagnosis: the effect of finite sample size

In computer-aided diagnosis (CAD), a frequently used approach for distinguishing normal and abnormal cases is first to extract potentially useful features for the classification task. Effective features are then selected from this entire pool of available features. Finally, a classifier is designed using the selected features. In this study, we investigated the effect of finite sample size on classification accuracy when classifier design involves stepwise feature selection in linear discriminant analysis, which is the most commonly used feature selection algorithm for linear classifiers. The feature selection and the classifier coefficient estimation steps were considered to be cascading stages in the classifier design process. We compared the performance of the classifier when feature selection was performed on the design samples alone and on the entire set of available samples, which consisted of design and test samples. The area Az under the receiver operating characteristic curve was used as our performance measure. After linear classifier coefficient estimation using the design samples, we studied the hold-out and resubstitution performance estimates. The two classes were assumed to have multidimensional Gaussian distributions, with a large number of features available for feature selection. We investigated the dependence of feature selection performance on the covariance matrices and means for the two classes, and examined the effects of sample size, number of available features, and parameters of stepwise feature selection on classifier bias. Our results indicated that the resubstitution estimate was always optimistically biased, except in cases where the parameters of stepwise feature selection were chosen such that too few features were selected by the stepwise procedure. When feature selection was performed using only the design samples, the hold-out estimate was always pessimistically biased. When feature selection was performed using the entire finite sample space, the hold-out estimates could be pessimistically or optimistically biased, depending on the number of features available for selection, the number of available samples, and their statistical distribution. For our simulation conditions, these estimates were always pessimistically (conservatively) biased if the ratio of the total number of available samples per class to the number of available features was greater than five.     

Physical exercise related improvement in obstructive sleep apnea. Look for the rostral fluid shift

Obstructive sleep apnea (OSA) is a common and underdiagnosed medical disorder. OSA is associated with the symptoms of excessive daytime sleepiness (EDS). These patients typically follow a sedentary lifestyle, and sedentary behavior is related to impaired fluid dynamics in the lower body, particularly the legs. In a supine position this fluid can move towards the neck, with a subsequent increase in upper airway (UA) resistance and UA collapse. Several studies have shown that rostral fluid shift worsens OSA; however, whether physical activity can influence this has not been tested.     

New tracheal sound feature for apnoea analysis

Sleep apnoea syndrome is common in the general population and is currently underdiagnosed. The aim of the present work was to develop a new tracheal sound feature for separation of apnoea events from non-apnoea time. Ten overnight recordings from apnoea patients containing 1,107 visually scored apnoea events totalling 7 h in duration and 72 h of non-apnoea time were included in the study. The feature was designed to describe the local spectral content of the sound signal. The median, maximum and mean smoothing of different time scales were compared in the feature extraction. The feature was designed to range from 0 to 1 irrespective of tracheal sound amplitudes. This constant range could offer application of the feature without patient-specific adjustments. The overall separation of feature values during apnoea events from non-apnoea time across all patients was good, reaching 80.8%. Due to the individual differences in tracheal sound signal amplitudes, developing amplitude-independent means for screening apnoea events is beneficial.     

Comparison of feature evaluation criteria for speech recognition based on electromyography

In this paper, we present a performance comparison of 14 feature evaluation criteria and 4 classifiers for isolated Thai word classification based on electromyography signals (EMG) to find a near-optimal criterion and classifier. Ten subjects spoke 11 Thai number words in both audible and silent modes while the EMG signal from five positions of the facial and neck muscles were captured. After signal collection and preprocessing, 22 EMG features widely used in the EMG recognition field were computed and were then evaluated based on 14 evaluation criteria including both independent criteria (IC) and dependent criteria (DC) for feature evaluation and selection. Subsequently, the top nine features were selected for each criterion, and were used as inputs to classifiers. Four types of classifier were employed with 10-fold cross-validation to estimate classification performance. The results showed that features selected with a DC on a Fisher’s least square linear discriminant classifier (D_FLDA) used with a linear Bayes normal classifier (LBN) gave the best average accuracies, of 93.25 and 80.12% in the audible and the silent modes, respectively.     

Assessment of Obstructive Sleep Apnea and its Severity during Wakefulness

In this article, a novel technique for assessment of obstructive sleep apnea (OSA) during wakefulness is proposed; the technique is based on tracheal breath sound analysis of normal breathing in upright sitting and supine body positions. We recorded tracheal breath sounds of 17 non-apneic individuals and 35 people with various degrees of severity of OSA in supine and upright sitting positions during both nose and mouth breathing at medium flow rate. We calculated the power spectrum, Kurtosis, and Katz fractal dimensions of the recorded signals and used the one-way analysis of variance to select the features, which were statistically significant between the groups. Then, the maximum relevancy minimum redundancy method was used to reduce the number of characteristic features to two. Using the best two selected features, we classified the participant into severe OSA and non-OSA groups as well as non-OSA or mild vs. moderate and severe OSA groups; the results showed more than 91 and 83% accuracy; 85 and 81% specificity; 92 and 95% sensitivity, for the two types of classification, respectively. The results are encouraging for identifying people with OSA and also prediction of OSA severity. Once verified on a larger population, the proposed method offers a simple and non-invasive screening tool for prediction of OSA during wakefulness.     

How does quantitative ultrasound compare to dual X-ray absorptiometry at various skeletal sites in relation to the WHO diagnosis categories?

The World Health Organisation (WHO) has proposed a set of guidelines for the diagnosis of osteoporosis in adult women based on a measurement of bone mineral density (BMD) expressed as the number of SD below young adult mean (t-score). In this study, we investigated the number of subjects classified as either osteopenic or osteoporotic according to these guidelines using dual X-ray absorptiometry (DXA), at the hip, at the spine and at the lower forearm and quantitative ultrasound (QUS), at the heel. A total of 247 men, 209 postmenopausal women and 195 premenopausal women were included in the study. Furthermore, the study provides the first normative data showing the influence of sex, age and menopause on broadband ultrasound attenuation (BUA) and speed of sound (SOS), as measured by the DTU-one imaging ultrasound scanner. The difference between the number of patients classified into either diagnosis group by the investigated parameters is large ranging from 25.9% of the women being diagnosed as osteopenic by BUA at the heel to 43.0% by BMD at the femoral neck. For men, the same range is from 20.5% by BUA to 44.1% by BMD at the femoral neck. For the classification into the osteoporotic group, the range is from 2.5% by intertrochanteric BMD to 24.4% by BMD at Ward's triangle for women and from 0% by SOS to 29.0% by BMD at Ward's triangle for men. Using total hip BMD as the reference parameter to categorize the subjects as normal, osteopenic or osteoporotic, the agreement of the other parameters with this classification is assessed in terms of sensitivity and specificity. We conclude that there are significant differences in the classification of osteoporosis/osteopenia depending on the site measured and the technique used for the bone mass assessment. Furthermore, we suggest that development of technique and site specific cut-off values may increase the accuracy of the classification of osteoporosis/osteopenia in both men and women.     

A wavelet-based optimal texture feature set for classification of brain tumours

In this work, the classification of brain tumours in magnetic resonance images is studied by using optimal texture features. These features are used to classify three sets of brain images—normal brain, benign tumour and malignant tumour. A wavelet-based texture feature set is derived from the region of interest. Each selected brain region of interest is characterized with both its energy and texture features extracted from the selected high frequency subband. An artificial neural network classifier is employed to evaluate the performance of these features. Feature selection is performed by a genetic algorithm. Principal component analysis and classical sequential methods are compared against the genetic approach in terms of the best recognition rate achieved and the optimal number of features. A classification performance of 98% is achieved in a genetic algorithm with only four of the available 29 features. Principal component analysis and classical sequential methods require a larger feature set to attain the similar classification accuracy of 98%. The optimal texture features such as range of angular second moment, range of sum variance, range of information measure of correlation II and energy selected by the genetic algorithm provide best classification performance with lower computational effort.     

A wavelet-based optimal texture feature set for classification of brain tumours

In this work, the classification of brain tumours in magnetic resonance images is studied by using optimal texture features. These features are used to classify three sets of brain images - normal brain, benign tumour and malignant tumour. A wavelet-based texture feature set is derived from the region of interest. Each selected brain region of interest is characterized with both its energy and texture features extracted from the selected high frequency subband. An artificial neural network classifier is employed to evaluate the performance of these features. Feature selection is performed by a genetic algorithm. Principal component analysis and classical sequential methods are compared against the genetic approach in terms of the best recognition rate achieved and the optimal number of features. A classification performance of 98% is achieved in a genetic algorithm with only four of the available 29 features. Principal component analysis and classical sequential methods require a larger feature set to attain the similar classification accuracy of 98%. The optimal texture features such as range of angular second moment, range of sum variance, range of information measure of correlation II and energy selected by the genetic algorithm provide best classification performance with lower computational effort.     

A comparison of different feature extraction methods for diagnosis of valvular heart diseases using PCG signals

This article presents a novel method for diagnosis of valvular heart disease (VHD) based on phonocardiography (PCG) signals. Application of the pattern classification and feature selection and reduction methods in analysing normal and pathological heart sound was investigated. After signal preprocessing using independent component analysis (ICA), 32 features are extracted. Those include carefully selected linear and nonlinear time domain, wavelet and entropy features. By examining different feature selection and feature reduction methods such as principal component analysis (PCA), genetic algorithms (GA), genetic programming (GP) and generalized discriminant analysis (GDA), the four most informative features are extracted. Furthermore, support vector machines (SVM) and neural network classifiers are compared for diagnosis of pathological heart sounds. Three valvular heart diseases are considered: aortic stenosis (AS), mitral stenosis (MS) and mitral regurgitation (MR). An overall accuracy of 99.47% was achieved by proposed algorithm.     

Automatic prediction of coronary artery disease from clinical narratives

Coronary Artery Disease (CAD) is not only the most common form of heart disease, but also the leading cause of death in both men and women (Coronary Artery Disease: MedlinePlus, 2015). We present a system that is able to automatically predict whether patients develop coronary artery disease based on their narrative medical histories, i.e., clinical free text. Although the free text in medical records has been used in several studies for identifying risk factors of coronary artery disease, to the best of our knowledge our work marks the first attempt at automatically predicting development of CAD. We tackle this task on a small corpus of diabetic patients. The size of this corpus makes it important to limit the number of features in order to avoid overfitting. We propose an ontology-guided approach to feature extraction, and compare it with two classic feature selection techniques. Our system achieves state-of-the-art performance of 77.4% F1 score.     

Temporal shifts in fluid in pulmonary hypertension with and without sleep apnea

Overnight extracellular rostral fluid shifts have been shown to be of importance in patients with fluid‐retaining states and are associated with a higher prevalence of sleep apnea. Pulmonary hypertension is frequently associated with right ventricular dysfunction and progressive right ventricular failure, and an increased prevalence of sleep apnea has been described. In light of the importance of fluid shifts in the pathophysiology of sleep apnea, we aimed to explore temporal fluid shifts in patients with pulmonary hypertension with and without sleep apnea. Patients with pulmonary hypertension (WHO Group 1 or 4) had overnight extracellular rostral fluid shift assessment before and a minimum of 3 months after initiation of pulmonary hypertension‐specific therapy. Fluid shift measurements of extracellular leg, abdominal, thoracic and neck fluid volumes were performed simultaneously. Twenty‐nine patients with pulmonary hypertension (age 55 ± 16 years, 69% female) participated. Sleep apnea was diagnosed in 15 subjects (apnea–hypopnea index 14 [8–27] per hr). There were no significant differences in baseline or overnight leg extracellular rostral fluid, abdominal extracellular rostral fluid, thoracic extracellular rostral fluid or neck extracellular rostral fluid between those with and without sleep apnea. There was a significant inverse correlation between the sleep apnea severity and the overnight change in leg extracellular rostral fluid (r = ?0.375, p = 0.049). There were no significant differences detected in overnight extracellular rostral fluid shifts from baseline to follow‐up. Treatment‐naïve patients with pulmonary hypertension both with and without sleep apnea demonstrate overnight extracellular rostral fluid shifts from the legs into the thorax and neck. Pulmonary hypertension‐specific treatment, while significantly improving cardiac haemodynamics, had little impact on nocturnal extracellular rostral fluid shifts or the presence of sleep apnea.     

Matrix decomposition based feature extraction for murmur classification

Heart murmurs often indicate heart valvular disorders. However, not all heart murmurs are organic. For example, musical murmurs detected in children are mostly innocent. Because of the challenges of mastering auscultation skills and reducing healthcare expenses, this study aims to discover new features for distinguishing innocent murmurs from organic murmurs, with the ultimate objective of designing an intelligent diagnostic system that could be used at home. Phonocardiographic signals that were recorded in an auscultation training CD were used for analysis. Instead of the discrete wavelet transform that has been used often in previous work, a continuous wavelet transform was applied on the heart sound data. The matrix that was derived from the continuous wavelet transform was then processed via singular value decomposition and QR decomposition, for feature extraction. Shannon entropy and the Gini index were adopted to generate features. To reduce the number of features that were extracted, the feature selection algorithm of sequential forward floating selection (SFFS) was utilized to select the most significant features, with the selection criterion being the maximization of the average accuracy from a 10-fold cross-validation of a classification algorithm called classification and regression trees (CART). An average sensitivity of 94%, a specificity of 83%, and a classification accuracy of 90% were achieved. These favorable results substantiate the effectiveness of the feature extraction methods based on the proposed matrix decomposition method.     

Effects of venous compression of the legs on overnight rostral fluid shift and obstructive sleep apnea

The amount of fluid displaced overnight from the legs into the neck as a consequence of lying recumbent correlates with the number of apneas and hypopneas per hour of sleep (AHI). Sedentary living promotes dependent fluid accumulation in the legs that can be counteracted by venous compression of the legs (compression stockings). We hypothesized that, in non-obese sedentary men with obstructive sleep apnea (OSA), wearing compression stockings during daytime will reduce the AHI by reducing the amount of fluid available for the displacement into the neck overnight. Polysomnography and measurement of overnight changes in leg fluid volume and neck circumference were performed at baseline and after one day of legs venous compression. The median AHI decreased from 30.9 (interquartile range 19.6-60.4) to 23.4 (12.9-31.8) (P=0.016) in association with a median 40% reduction in the change in leg fluid volume (P=0.016) and a median 42% reduction in the increase in neck circumference (P=0.016). These results provide proof-of-principle that overnight fluid displacement into the neck plays a causative role in OSA.     

基于机器学习的心音识别分类研究

心音信号可反映心脏的病理信息,是诊断心脏健康的重要依据之一。本文首先从心音信号提取时频域、梅尔倒谱系数等145个特征作为机器学习的输入数据集,然后在随机森林、LightGBM、XGBoost、GBDT、SVM共5种分类器中选出效果最佳分类器与递归特征消除算法结合进行数据挖掘,找出重要特征集并对其分类效果做比较与分析,最后运用Stacking模型融合方法优化模型。数据挖掘特征子集比同数量特征子集在准确率、召回率、精确率、F1值上分别提高了33.51%、14.54%、20.61%、24.04%;采用LightGBM和SVM模型融合可将F1值提高至92.6%。本文提出了一种有效的心音识别分类方法,挖掘出心音最重要的8个特征,为临床诊断提供参考。     

Influence of sex and age on duration and frequency of sleep apnea events

OBJECTIVES: Differences between men and women potentially provide insight into the regulation sleep apnea events. This study, therefore, examined how apnea frequency and duration varied according to age, sex, and sleep stage in a clinical population. DESIGN: NA SETTING: NA PATIENTS: Patients were 215 women and 215 men referred to a sleep disorders center with symptoms of obstructive sleep apnea and matched for BMI. Apnea events were compared across three age groups (18-39, 40-59, and 60-88 years) in stage 2 and in REM sleep. INTERVENTIONS: NA RESULTS: In stage 2 sleep, young and middle aged women were similar averaging 15 and 13 apnea events per hour respectively. Men had significantly more events averaging 27 and 30 events per hour for the corresponding age groups. The apnea frequency doubled from middle age to older women, and the sex difference narrowed between the older males and females to a non significant difference (26 events per hour for women versus 34 events per hour for men). Apnea duration was significantly longer in men than in women. Stage 2 apnea duration increased significantly with age for men (20.1, 21.5, 23.8 s) and women (16.7, 18.3, 20.6 s) across the three age groups. This also occurred in REM sleep in for men (22.8, 26.5, 29.8 s) and women (19.3, 22.4, 26.6 s). CONCLUSIONS: Duration did not demonstrate the marked "menopausal effect" that there was for apnea frequency. Female gender and younger age conferred benefit primarily by preventing airway collapse (reduced apnea frequency) with less of an effect on apnea duration, i.e., the ability to end the apnea. Compared to stage 2 sleep, REM sleep reduced the differences between men and women in apnea frequency. One explanation may be that differences in muscle tone of the upper airway account for the sex differences in apnea frequency.     

Sex Differences in the Association of Regional Fat Distribution with the Severity of Obstructive Sleep Apnea

Study Objectives:

To describe sex differences in the associations between severity of obstructive sleep apnea (OSA) and measures of obesity in body regions defined using both dual-energy absorptiometry and traditional anthropometric measures in a sleep-clinic sample.

Design:

A prospective case-series observational study.

Setting:

The Western Australian Sleep Health Study operating out of the Sir Charles Gairdner Hospital Sleep Clinic, Perth, Western Australia.

Participants:

Newly referred clinic patients (60 men, 36 women) suspected of having OSA.

Interventions:

N/A

Measurements and Results:

Obstructive sleep apnea severity was defined by apnea-hypopnoea index from laboratory-based overnight polysomnography. Body mass index, neck, waist and hip circumference, neck-to-waist ratio, and waist-to-hip ratio were measured. Dual energy absorptiometry measurements included percentage fat and lean tissue. Multivariate regression models for each sex were developed. In women, percentage of fat in the neck region and body mass index together explained 33% of the variance in apnea-hypopnea index. In men, percentage of fat in the abdominal region and neck-to-waist ratio together accounted for 37% of the variance in apnea-hypopnea index.

Conclusions:

Regional obesity is associated with obstructive sleep apnea severity, although differently in men and women. In women, a direct influence of neck fat on the upper airway patency is implicated. In men, abdominal obesity appears to be the predominant influence. The apnea-hypopnea index was best predicted by a combination of Dual Energy Absorptiometry-measured mass and traditional anthropometric measurements.

Citation:

Simpson L; Mukherjee S; Cooper MN; Ward KL; Lee JD; Fedson AC; Potter J; Hillman Fanzca DR; Eastwood P; Palmer LJ; Kirkness J. Sex differences in the association of regional fat distribution with the severity of obstructive sleep apnea. SLEEP 2010;33(4):467-474