Jordi Sola-Soler

Automatic detection of snoring signals: validation with simple snorers and OSAS patients

Relationship between snoring and Obstructive Sleep Apnea Syndrome (OSAS) has been reported in the literature. Recently, studies of snoring sound intensity, but also estimation of spectral features for each snoring episode, have been published. Usually, patients that are suspected of OSAS pathology are studied by polysomnography during all the night. To analyze the snoring signal, it is very useful to automatically detect each episode, in order to calculate several features that describe the signal. In this work an automatic detection algorithm of acoustic snoring signals has been designed, to work with long duration respiratory sound recordings. Two blocs compose the detector. The former is a segmentation subsystem that detects changes of variance on the signal. The latter is a 2-layer Feedforward Multilayer Neural Network with backpropagation learning algorithm. The network was trained with 625-selected events, including snores with different shapes and characteristics, from normal snorers and OSAS patients, and other sounds. In this way, the detector was designed to select snoring episodes from simple snorers and OSAS patients, and to reject cough, voice and other artifacts. The detector has been applied to real snoring signals recorded during polysomnographic studies. In order to validate the detector, more than 500 snores were analyzed from 10 excerpts, taken at random from a database of 30 snorer subjects with different apnea/hipoanea index (AHI). Results were compared with manual annotations done by a medical doctor. The detector showed a good performance and achieved a Sensitivity of 82% and a Positive Predictive Value of 90%.

Automatic classification of subjects with and without Sleep Apnea through snoring analysis

A new method for indirect identification of Sleep Apnea patients through snoring characteristics is proposed. The method uses a logistic regression model which is fed with several time and frequency parameters from snores and their variability. The information is contained in all the snores automatically detected in nocturnal sound recordings. In the validation of the model, subjects are classified with a sensitivity higher than 93% and a specificity between 73% and 88% when all detected snores are used. The model can also be adjusted to obtain 100% specificity with a corresponding sensitivity between 70% and 87%. This results are better than previous reported methods based on snoring analysis, but with a single channel, and are comparable to the classification scores of several portable apnea monitors when evaluated on a similar number of patients. This technique is a promising tool for the screening of snorers, allowing snorers with a low apnea-hypopnea index (AHI< 10) to avoid a full-night polysomnographic study at the hospital.

Automatic snoring signal analysis in sleep studies

Snoring has been related to vibration of upper airway during sleep. It has been reported in the literature as a risk factor of different diseases, such as obstructive sleep apnea syndrome (OSAS) and other breathing abnormalities during sleep. Recently, our group has developed an automatic detector of snores to be applied in long-term sleep studies. This detector includes segmentation and classification blocs, based on a feedforward multilayer neural network. In this work, a complete procedure for detector validation is proposed, including annotation of different episodes: snores, sounds during inspiration and exhalation, speech and noise artifacts. A database of 948 episodes was manually annotated by a medical doctor in respiratory sound signals from 8 male subjects (4 normal snorers and 4 OSAS patients). The ratio non-snores/total annotated episodes was 53%. The detector shown a good performance, obtaining a sensitivity of 76,1%, a positive predictive value of 75,6% and a specificity of 82,8%. The automatic detector was applied to 6-hour snoring signals, corresponding to 37 subjects (12 females/25 males, 20 snorers/17 OSAS). Significant results shown differences between snorers and OSAS patients, and suggest that snore variability could be higher in OSAS patients.

Formant frequencies of normal breath sounds of snorers may indicate the risk of Obstructive Sleep Apnea Syndrome

Several differences between the airway of normal subjects and those with OSAS are well known. The characteristics of the upper airway may be indirectly studied through the formant frequencies of breathing sounds. In this work we analyze the formants of inspiration and exhalation sounds in snoring subjects with and without OSAS. Formant frequencies of inspiration and exhalation appear in the same bands as snores. Formant F1 is significantly lower in inspiration episodes of OSAS patients (p=0.008) with a decreasing tendency as the AHI increases (r=−0.705). In addition, this formant has a significantly higher variability SF1 in pathological subjects, for both inspiration (p=0.022) and exhalation (p=0.038) episodes, as was previously found in snores. A higher variability of formant frequencies seems to be an indicator of the presence of OSAS. The proposed technique could allow the identification of OSAS patients from normal breathing alone.

Identification of Obstructive Sleep Apnea patients from tracheal breath sound analysis during wakefulness in polysomnographic studies

Obstructive Sleep Apnea (OSA) is currently diagnosed by a full nocturnal polysomnography (PSG), a very expensive and time-consuming method. In previous studies we were able to distinguish patients with OSA through formant frequencies of breath sound during sleep. In this study we aimed at identifying OSA patients from breath sound analysis during wakefulness. The respiratory sound was acquired by a tracheal microphone simultaneously to PSG recordings. We selected several cycles of consecutive inspiration and exhalation episodes in 10 mild-moderate (AHI<;30) and 13 severe (AHI>=30) OSA patients during their wake state before getting asleep. Each episodes formant frequencies were estimated by linear predictive coding. We studied several formant features, as well as their variability, in consecutive inspiration and exhalation episodes. In most subjects formant frequencies were similar during inspiration and exhalation. Formant features in some specific frequency band were significantly different in mild OSA as compared to severe OSA patients, and showed a decreasing correlation with OSA severity. These formant characteristics, in combination with some anthropometric measures, allowed the classification of OSA subjects between mild-moderate and severe groups with sensitivity (specificity) up to 88.9% (84.6%) and accuracy up to 86.4%. In conclusion, the information provided by formant frequencies of tracheal breath sound recorded during wakefulness may allow identifying subjects with severe OSA.

Snoring analysis for the screening of sleep apnea hypopnea syndrome with a single-channel device developed using polysomnographic and snoring databases

Several studies have shown differences in acoustic snoring characteristics between patients with Sleep Apnea-Hypopnea Syndrome (SAHS) and simple snorers. Usually a few manually isolated snores are analyzed, with an emphasis on postapneic snores in SAHS patients. Automatic analysis of snores can provide objective information over a longer period of sleep. Although some snore detection methods have recently been proposed, they have not yet been applied to full-night analysis devices for screening purposes. We used a new automatic snoring detection and analysis system to monitor snoring during full-night studies to assess whether the acoustic characteristics of snores differ in relation to the Apnea-Hypopnea Index (AHI) and to classify snoring subjects according to their AHI. A complete procedure for device development was designed, using databases with polysomnography (PSG) and snoring signals. This included annotation of many types of episodes by an expert physician: snores, inspiration and exhalation breath sounds, speech and noise artifacts, The AHI of each subject was estimated with classical PSG analysis, as a gold standard. The system was able to correctly classify 77% of subjects in 4 severity levels, based on snoring analysis and sound-based apnea detection. The sensitivity and specificity of the system, to identify healthy subjects from pathologic patients (mild to severe SAHS), were 83% and 100%, respectively. Besides, the Apnea Index (AI) obtained with the system correlated with the obtained by PSG or Respiratory Polygraphy (RP) (r=0.87, p<0.05).

Cardiorespiratory Phase Synchronization in OSA subjects during wake and sleep states

Cardiorespiratory Phase Synchronization (CRPS) is a manifestation of coupling between cardiac and respiratory systems complementary to Respiratory Sinus Arrhythmia. In this work, we investigated CRPS during wake and sleep stages in Polysomnographic (PSG) recordings of 30 subjects suspected from Obstructive Sleep Apnea (OSA). The population was classified into three severity groups according to the Apnea Hypopnea Index (AHI): G1 (AHI<;15), G2 (15<;=AHI<;30) and G3 (AHI>30). The synchrogram between single lead ECG and respiratory abdominal band signals from PSG was computed with the Hilbert transform technique. The different phase locking ratios (PLR) m:n were monitored throughout the night. Ratio 4:1 was the most frequent and it became more dominant as OSA severity increased. CRPS was characterized by the percentage of synchronized time (%Sync) and the average duration of synchronized epochs (AvDurSync) using three different thresholds. Globally, we observed that %Sync significantly decreased and AvDurSync slightly increased with OSA severity. A high synchronization threshold enhanced these population differences. %Sync was significantly higher in NREM than in REM sleep in G2 and G3 groups. Population differences observed during sleep did not translate to the initial wake state. Reduced CRPS could be an early marker of OSA severity during sleep, but further studies are needed to determine whether CRPS is also present during wakefulness.

Regular and non regular snore features as markers of SAHS

Sleep Apnea-Hypopnea Syndrome (SAHS) diagnosis is still done with an overnight multi-channel polysomnography. Several efforts are being made to study profoundly the snore mechanism and discover how it can provide an opportunity to diagnose the disease. This work introduces the concept of regular snores, defined as the ones produced in consecutive respiratory cycles, since they are produced in a regular way, without interruptions. We applied 2 thresholds (THadaptive and THmedian) to the time interval between successive snores of 34 subjects in order to select regular snores from the whole all-night snore sequence. Afterwards, we studied the effectiveness that parameters, such as time interval between successive snores and the mean intensity of snores, have on distinguishing between different levels of SAHS severity (AHI (Apnea-Hypopnea Index)<5h−1, AHI<10 h−1, AHI<15h−1, AHI<30h−1). Results showed that THadaptive outperformed THmedian on selecting regular snores. Moreover, the outcome achieved with non-regular snores intensity features suggests that these carry key information on SAHS severity.

Bayes classification of snoring subjects with and without Sleep Apnea Hypopnea Syndrome, using a Kernel method

The gold standard for diagnosing Sleep Apnea Hypopnea Syndrome (SAHS) is the Polysomnography (PSG), an expensive, labor-intensive and time-consuming procedure. It would be helpful to have a simple screening method that allowed to early determining the severity of a subject prior to his/her enrolment for a PSG. Several differences have been reported in the acoustic snoring characteristics between simple snorers and SAHS patients. Previous studies usually classify snoring subjects into two groups given a threshold of Apnea-Hypoapnea Index (AHI). Recently, Bayes multi-group classification with Gaussian Probability Density Function (PDF) has been proposed, using snore features in combination with apnea-related information. In this work we show that the Bayes classifier with Kernel PDF estimation outperforms the Gaussian approach and allows the classification of SAHS subjects according to their severity, using only the information obtained from snores. This could be the base of a single channel, snore-based, screening procedure for SAHS.