Characterizing the NREM/REM sleep specific obstructive sleep apnea severity using snore sounds.

Obstructive Sleep Apnea (OSA) patients have frequent breathing obstructions and upper airway (UA) collapse during sleep. It is clinically important to estimate OSA severity separately for Rapid Eye Movement (REM) and non-REM (NREM) sleep states, but the task requires Polysomnography (PSG) which uses about 15-20 body contact sensors and subjective assessment. Almost all OSA patients snore. Vibration in narrowed UA muscles cause snoring in OSA. Moreover, as sleep states are associated with distinct breathing patterns and UA muscle tone, REM/NREM specific information must be available via snore/breathing sounds. Our previous works have shown that snoring carries significant information related to REM/NREM sleep states and OSA. We hypothesized that such information from snoring sound could be used to characterize OSA specific to REM/NREM sleep states independent of PSG. We acquired overnight audio recording from 91 patients (56 males and 35 females) undergoing PSG and labeled snore sounds as belonging to REM/NREM stages based on PSG. We then developed features to capture REM/NREM specific information and trained logistic regression (LR) classifier models to map snore features to OSA severity bands. Considering separate LR models for males and females, we achieved 94-100% sensitivity (84-89% specificity) for NREM stages at the OSA severity threshold of 30 events/h. Corresponding sensitivity for REM stages were 92-97% with specificity 83-85%. Results indicate that it is feasible to estimate severe/non-severe OSA in REM/NREM sleep based on snore/breathing sounds alone, acquired using simple bedside sound acquisition devices such as mobile phones.

Variations of snoring properties with macro sleep stages in a population of Obstructive Sleep Apnea patients.

Snoring is common in Obstructive Sleep Apnea (OSA) patients. Snoring originates from the vibration of soft tissues in the upper airways (UA). Frequent UA collapse in OSA patients leads to sleep disturbances and arousal. In a routine sleep diagnostic procedure, sleep is broadly divided into rapid eye movement (REM), non-REM (NREM) states. These Macro-Sleep States (MSS) are known to be involved with different neuromuscular activities. These differences should influence the UA mechanics in OSA patients as well as the snoring sound (SS). In this paper, we propose a logistic regression model to investigate whether the properties of SS from OSA patients can be separated into REM/NREM group. Analyzing mathematical features of more than 500 SS events from 7 OSA patients, the model achieved 76% (± 0.10) sensitivity and 75% (± 0.10) specificity in categorizing REM and NREM related snores. These results indicate that snoring is affected by REM/NREM states and proposed method has potential in differentiating MSS.

Neural network based algorithm for automatic identification of cough sounds.

Cough is the most common symptom of the several respiratory diseases containing diagnostic information. It is the best suitable candidate to develop a simplified screening technique for the management of respiratory diseases in timely manner, both in developing and developed countries, particularly in remote areas where medical facilities are limited. However, major issue hindering the development is the non-availability of reliable technique to automatically identify cough events. Medical practitioners still rely on manual counting, which is laborious and time consuming. In this paper we propose a novel method, based on the neural network to automatically identify cough segments, discarding other sounds such a speech, ambient noise etc. We achieved the accuracy of 98% in classifying 13395 segments into two classes, 'cough' and 'other sounds', with the sensitivity of 93.44% and specificity of 94.52%. Our preliminary results indicate that method can develop into a real-time cough identification technique in continuous cough monitoring systems.

Cough sound analysis - a new tool for diagnosing pneumonia.

Pneumonia kills over 1,800,000 children annually throughout the world. Prompt diagnosis and proper treatment are essential to prevent these unnecessary deaths. Reliable diagnosis of childhood pneumonia in remote regions is fraught with difficulties arising from the lack of field-deployable imaging and laboratory facilities as well as the scarcity of trained community healthcare workers. In this paper, we present a pioneering class of enabling technology addressing both of these problems. Our approach is centered on automated analysis of cough and respiratory sounds, collected via microphones that do not require physical contact with subjects. We collected cough sounds from 91 patients suspected of acute respiratory illness such as pneumonia, bronchiolitis and asthma. We extracted mathematical features from cough sounds and used them to train a Logistic Regression classifier. We used the clinical diagnosis provided by the paediatric respiratory clinician as the gold standard to train and validate our classifier against. The methods proposed in this paper could separate pneumonia from other diseases at a sensitivity and specificity of 94% and 75% respectively, based on parameters extracted from cough sounds alone. Our method has the potential to revolutionize the management of childhood pneumonia in remote regions of the world.

Obstructive sleep apnea screening by integrating snore feature classes.

Obstructive sleep apnea (OSA) is a serious sleep disorder with high community prevalence. More than 80% of OSA suffers remain undiagnosed. Polysomnography (PSG) is the current reference standard used for OSA diagnosis. It is expensive, inconvenient and demands the extensive involvement of a sleep technologist. At present, a low cost, unattended, convenient OSA screening technique is an urgent requirement. Snoring is always almost associated with OSA and is one of the earliest nocturnal symptoms. With the onset of sleep, the upper airway undergoes both functional and structural changes, leading to spatially and temporally distributed sites conducive to snore sound (SS) generation. The goal of this paper is to investigate the possibility of developing a snore based multi-feature class OSA screening tool by integrating snore features that capture functional, structural, and spatio-temporal dependences of SS. In this paper, we focused our attention to the features in voiced parts of a snore, where quasi-repetitive packets of energy are visible. Individual snore feature classes were then optimized using logistic regression for optimum OSA diagnostic performance. Consequently, all feature classes were integrated and optimized to obtain optimum OSA classification sensitivity and specificity. We also augmented snore features with neck circumference, which is a one-time measurement readily available at no extra cost. The performance of the proposed method was evaluated using snore recordings from 86 subjects (51 males and 35 females). Data from each subject consisted of 6-8 h long sound recordings, made concurrently with routine PSG in a clinical sleep laboratory. Clinical diagnosis supported by standard PSG was used as the reference diagnosis to compare our results against. Our proposed techniques resulted in a sensitivity of 93±9% with specificity 93±9% for females and sensitivity of 92±6% with specificity 93±7% for males at an AHI decision threshold of 15 events/h. These results indicate that our method holds the potential as a tool for population screening of OSA in an unattended environment.

Impact of gender on snore-based obstructive sleep apnea screening.

Obstructive sleep apnea syndrome (OSA) is a serious widespread disease in which upper airways (UA) are collapsed during sleep. OSA has marked male predominance in prevalence. Although women are less vulnerable to OSA, under-diagnosed OSA in women may associate with serious consequences. Snoring is commonly associated with OSA and one of the earliest symptoms. Snore sounds (SS) are generated due to vibration of the collapsing soft tissues of the UA. Structural and functional properties of the UA are gender dependent. SS capture these time varying gender attributed UA properties and those could be embedded in the acoustic properties of SS. In this paper, we investigate the gender-specific acoustic property differences of SS and try to exploit these differences to enhance the snore-based OSA detection performance. We developed a snore-based multi-feature vector for OSA screening and one time-measured neck circumference was augmented. Snore features were estimated from SS recorded in a sleep laboratory from 35 females and 51 males and multi-layer neural network-based pattern recognition algorithms were used for OSA/non-OSA classification. The results were K-fold cross-validated. Gender-dependent modeling resulted in an increase of around 7% in sensitivity and 6% in specificity at the decision threshold AHI = 15 against a gender-neutral model. These results established the importance of adopting gender-specific models for the snore-based OSA screening technique.

Automated algorithm for Wet/Dry cough sounds classification.

Cough is the most common symptom of several respiratory diseases. It is a defense mechanism of the body to clear the respiratory tract from foreign materials inhaled accidentally or produced internally by infections. The identification of wet and dry cough is an important clinical finding, aiding in the differential diagnosis. Wet coughs are more likely to be associated with bacterial infections. At present, the wet/dry decision is based on the subjective judgment of a physician, during a typical consultation session. It is not available for long term monitoring or in the assessment of treatment efficacy. In this paper we address these issues and develop fully automated technology to classify cough into 'Wet' and 'Dry' categories. We propose novel features and a Logistic regression-based model for the classification of coughs into wet/dry classes. The performance of the method was evaluated on a clinical database of pediatric and adult coughs recorded using a bed-side non-contact microphone. The sensitivity and specificity of the classification were obtained as 79±9% and 72.7±8.7% respectively. These indicate the potential of the method as a useful clinical tool for cough monitoring, especially at home settings.

High frequency analysis of cough sounds in pediatric patients with respiratory diseases.

Cough is a common symptom in a range of respiratory diseases and is considered a natural defense mechanism of the body. Despite its critical importance in the diagnosis of illness, there are no golden methods to objectively assess cough. In a typical consultation session, a physician may briefly listen to the cough sounds using a stethoscope placed against the chest. The physician may also listen to spontaneous cough sounds via naked ears, as they naturally propagate through air. Cough sounds carry vital information on the state of the respiratory system but the field of cough analysis in clinical medicine is in its infancy. All existing cough analysis approaches are severely handicapped by the limitations of the human hearing range and simplified analysis techniques. In this paper, we address these problems, and explore the use of frequencies covering a range well beyond the human perception (up to 90 kHz) and use wavelet analysis to extract diagnostically important information from coughs. Our data set comes from a pediatric respiratory ward in Indonesia, from subjects diagnosed with asthma, pneumonia and rhinopharyngitis. We analyzed over 90 cough samples from 4 patients and explored if high frequencies carried useful information in separating these disease groups. Multiple regression analysis resulted in coefficients of determination (R(2)) of 77-82% at high frequencies (15 kHz-90 kHz) indicating that they carry useful information. When the high frequencies were combined with frequencies below 15kHz, the R(2) performance increased to 85-90%.

High frequency region of the snore spectra carry important information on the disease of sleep apnoea.

Snoring is the most common symptom of obstructive sleep apnoea (OSA). Several researchers have reported differences between the power spectra of non-OSA and OSA snorers. The traditional approach over the years has been to record snore sounds at a bandwidth of < 5 kHz. Narrowing of the upper airways during OSA events and the resulting upward shift of snore frequencies also lend support to the idea of examining snore sounds beyond 5 kHz. In this paper, we compute the power spectra of snores in three different bands defined as: low-frequency band (LFB: < 5 kHz); middle-frequency band (MFB: 5-10 kHz) and high-frequency band (HFB: 10-20 kHz). We illustrate that there is a significant difference between non-OSA snorers (Apnoea Hypopnoea Index (AHI) < 10) and OSA snorers (AHI > 10) in the region > 5 kHz. We then develop a feature to diagnose OSA based on the spectral differences in the high frequency region and evaluate its performance on a database of 20 subjects. Our results strongly suggest that the high-frequency region of the snore sounds carry information, hitherto disregarded, on the disease of sleep apnoea.

A method to screen obstructive sleep apnea using multi-variable non-intrusive measurements.

Obstructive sleep apnea (OSA) is a serious sleep disorder. The current standard OSA diagnosis method is polysomnography (PSG) testing. PSG requires an overnight hospital stay while physically connected to 10-15 channels of measurement. PSG is expensive, inconvenient and requires the extensive involvement of a sleep technologist. As such, it is not suitable for community screening. OSA is a widespread disease and more than 80% of sufferers remain undiagnosed. Simplified, unattended and cheap OSA screening methods are urgently needed. Snoring is commonly associated with OSA but is not fully utilized in clinical diagnosis. Snoring contains pseudo-periodic packets of energy that produce characteristic vibrating sounds familiar to humans. In this paper, we propose a multi-feature vector that represents pitch information, formant information, a measure of periodic structure existence in snore episodes and the neck circumference of the subject to characterize OSA condition. Snore features were estimated from snore signals recorded in a sleep laboratory. The multi-feature vector was applied to a neural network for OSA/non-OSA classification and K-fold cross-validated using a random sub-sampling technique. We also propose a simple method to remove a specific class of background interference. Our method resulted in a sensitivity of 91 ± 6% and a specificity of 89 ± 5% for test data for AHI(THRESHOLD) = 15 for a database consisting of 51 subjects. This method has the potential as a non-intrusive, unattended technique to screen OSA using snore sound as the primary signal.

A new measure to quantify sleepiness using higher order statistical analysis of EEG.

Chronic sleepiness is a common symptom in the sleep disorders, such as, Obstructive Sleep Apnea, Periodic leg movement syndrome, narcolepsy etc. It affects 5% of the adult population and is associated with significant morbidity and increased risk to individual and society. MSLT and MWT are the existing tests for measuring sleepiness. Sleep Latency (SL) is the main measures of sleepiness computed in these tests. Existing method of SL computation relies on the visual extraction of specific features in multi-channel electrophysiological data (EEG, EOG, and EMG) using the R&K criteria (1968). This process is cumbersome, time consuming, and prone to inter and intra-scorer variability. In this paper we propose a fully automated, objective sleepiness analysis technique based on the single channel of EEG. The method uses a one-dimensional slice of the EEG Bisprectrum representing a nonlinear transformation of the underlying EEG generator to compute a novel index called Sleepiness Index. The SL is then computed from the SI. A strong correlation (r = 0.93, rho = 0.0001) was found between technician scored SL and that computed via SI. The proposed Sleepiness Index can provide an elegant solution to the problems surrounding manual scoring and objective sleepiness.

Normal probability testing of snore signals for diagnosis of obstructive sleep apnea.

Obstructive Sleep Apnea (OSA) is a highly prevalent disease in which upper airways are collapsed during sleep, leading to serious consequences. The standard method of OSA diagnosis is known as Polysomnography (PSG), which requires an overnight stay in a specifically equipped facility, connected to over 15 channels of measurements. PSG requires (i) contact instrumentation and, (ii) the expert human scoring of a vast amount of data based on subjective criteria. PSG is expensive, time consuming and is difficult to use in community screening or pediatric assessment. Snoring is the most common symptom of OSA. Despite the vast potential, however, it is not currently used in the clinical diagnosis of OSA. In this paper, we propose a novel method of snore signal analysis for the diagnosis of OSA. The method is based on a novel feature that quantifies the non-Gaussianity of individual episodes of snoring. The proposed method was evaluated using overnight clinical snore sound recordings of 86 subjects. The recordings were made concurrently with routine PSG, which was used to establish the ground truth via standard clinical diagnostic procedures. The results indicated that the developed method has a detectability accuracy of 97.34%.

Modelling of polysomnographic respiratory measurements for artefact detection and signal restoration.

Polysomnography (PSG), which incorporates measures of sleep with measures of EEG arousal, air flow, respiratory movement and oxygenation, is universally regarded as the reference standard in diagnosing sleep-related respiratory diseases such as obstructive sleep apnoea syndrome. Over 15 channels of physiological signals are measured from a subject undergoing a typical overnight PSG session. The signals often suffer from data losses, interferences and artefacts. In a typical sleep scoring session, artefact-corrupted signal segments are visually detected and removed from further consideration. This is a highly time-consuming process, and subjective judgement is required for the job. During typical sleep scoring sessions, the target is the detection of segments of diagnostic interest, and signal restoration is not utilized for distorted segments. In this paper, we propose a novel framework for artefact detection and signal restoration based on the redundancy among respiratory flow signals. We focus on the air flow (thermistor sensors) and nasal pressure signals which are clinically significant in detecting respiratory disturbances. The method treats the respiratory system and other organs that provide respiratory-related inputs/outputs to it (e.g., cardiovascular, brain) as a possibly nonlinear coupled-dynamical system, and uses the celebrated Takens embedding theorem as the theoretical basis for signal prediction. Nonlinear prediction across time (self-prediction) and signals (cross-prediction) provides us with a mechanism to detect artefacts as unexplained deviations. In addition to detection, the proposed method carries the potential to correct certain classes of artefacts and restore the signal. In this study, we categorize commonly occurring artefacts and distortions in air flow and nasal pressure measurements into several groups and explore the efficacy of the proposed technique in detecting/recovering them. The results we obtained from a database of clinical PSG signals indicated that the proposed technique can detect artefacts/distortions with a sensitivity>88.3% and specificity>92.4%. This work has the potential to simplify the work done by sleep scoring technicians, and also to improve automated sleep scoring methods.

Feature extraction for snore sound via neural network processing.

Snore sound (SS) is the earliest and the most common symptom of Obstructive Sleep Apnea (OSA) which is a serious disease caused by the collapse of upper airways during sleep. SS should carry vital information on the state of the upper airways and is simple to acquire and rich in features but their analysis is complicated. In this study we use neural network (NN) based method to model SS via a simple second order one-step predictor. We show that the some hidden information/feature of a SS can be conveniently captured in the connection-weight-space (CWS) of the NN, after a process of supervised training. The availability of the proposed method is investigated by performing independent component analysis (ICA) on CWS.

Tracking the states of a nonlinear and nonstationary system in the weight-space of artificial neural networks.

We propose a novel interpretation and usage of Neural Network (NN) in modeling physiological signals, which are allowed to be nonlinear and/or nonstationary. The method consists of training a NN for the k-step prediction of a physiological signal, and then examining the connection-weight-space (CWS) of the NN to extract information about the signal generator mechanism. We define a novel feature, Normalized Vector Separation (gamma(ij)), to measure the separation of two arbitrary states "i" and "j" in the CWS and use it to track the state changes of the generating system. The performance of the method is examined via synthetic signals and clinical EEG. Synthetic data indicates that gamma(ij) can track the system down to a SNR of 3.5 dB. Clinical data obtained from three patients undergoing carotid endarterectomy of the brain showed that EEG could be modeled (within a root-means-squared-error of 0.01) by the proposed method, and the blood perfusion state of the brain could be monitored via gamma(ij), with small NNs having no more than 21 connection weight altogether.

EEG source localization: comparative study of classical and neural network methods.

We address the problem of estimating biopotential sources within the brain, based on EEG signals observed on the scalp. This problem, known as the inverse problem of electrophysiology, has no closed-form solution, and requires iterative techniques such as the Levenberg-Marquardt (LM) algorithm. Considering the nonlinear nature of the inverse problem, and the low signal to noise ratio inherent in EEG signals, a backpropagation neural network (BPN) has been recently proposed as a solution. The technique has not been properly compared with classical techniques such as the LM method, or with more recent neural network techniques such as the Radial Basis Function (RBF) network. In this paper, we provide improved strategies based on BPN and consider RBF networks in solving the inverse problem. We compare the performances of BPN, RBF and a hybrid technique with that of the classical LM method.

RBF networks for source localization in quantitative electrophysiology.

The backpropagation neural network methods have been proposed recently to solve the inverse problem in quantitative electrophysiology. A major advantage of the technique is that once a neural network is trained, it no longer requires iterations or access to sophisticated computations. We propose to use RBF networks for source localization in the brain, and systematically compare their performance to those of Levenberg-Marquardt (LM) algorithms. We show the use of two types of Radial Basis Function Networks (RBF) network: a classic network with fixed number of hidden layer neurons and an improved network, Minimal Resource Allocation Network (MRAN), recently proposed by one of the authors, capable for dynamically configuring its structure so as to obtain a compact topology to match the data presented to it.

A novel scheme for the selective stimulation of nerve fibers using nonlinear properties of action potentials.

In this paper, we propose a novel technique for selective stimulation of nerve fibers. We show that a set of point electrodes arranged in the 3-D space around a nerve trunk can be used to systematically synthesize highly useful activation patterns within the nerve, by exploiting the spatial arrangement of the electrodes and the excitation currents. Using such activation patterns, we present a novel scheme to selectively stimulate nerve fibers, based on the nonlinear properties of action potential generation. We illustrate the developed techniques via computer simulations of a nerve trunk consisting of a large number of nerve fibers. The results indicate that the proposed technique has great potential to achieve position selective stimulation of nerve in FES.

Spectral information changes in obtaining heart rate variability from tachometer R-R interval signals.

The spontaneous changes in the heartbeat provide valuable information regarding serious cardiac pathologies such as Sudden Cardiac Death. Subtle anomalies in the heart rate can be discovered via an analysis of the Heart Rate Variability (HRV) signal, a sequence of numbers representing the instantaneous heart rate over time. The HRV signal is commonly modeled by an integral pulse frequency modulation (IPFM) scheme. Based on the model, several techniques have been proposed in the literature to capture features of the HRV signal. They however, suffer from severe spurious peaks, especially when the HRV signals contain multiple spectral lines. In this paper, we propose a new method to minimize spurious spectral peaks via the technique of phase interpolation.

Wavelet transforms in estimating scatterer spacing from ultrasound echoes.

Ultrasound echoes from organs such as the liver display resolvable periodicity due to regular scattering centers within tissue. The spacing among such scattering centers has been proposed as a signature to characterize diffuse and focal diseases of the liver. Even though it is highly desirable to be able to estimate an inter-scatterer-spacing (ISS) distribution, current methods can estimate only the mean value of scatterer spacing (MSS) over a tissue length. In this paper, we propose a wavelet transform-based technique that is capable of estimating the location of each scattering center, making it possible to obtain the ISS distribution. We represent liver tissue with a point scatterer model, and show, via computer simulations, that the use of multi-scale information in the wavelet scale-space allows us to estimate the locations of regular scattering centers. We show that both the observation noise and random ultrasound returns from unresolvable tissue microstructure can be removed successfully in the wavelet domain via the properties of the modulus maxima sequence of observation across different scales.