Detection of breathing sounds during sleep using non-contact audio recordings

Abstract

Evaluation of respiratory activity during sleep is essential in order to reliably diagnose sleep disorder breathing (SDB); a condition associated with serious cardio-vascular morbidity and mortality. In the current study, we developed and validated a robust automatic breathing-sounds (i.e. inspiratory and expiratory sounds) detection system of audio signals acquired during sleep. Random forest classifier was trained and tested using inspiratory/expiratory/noise events (episodes), acquired from 84 subjects consecutively and prospectively referred to SDB diagnosis in sleep laboratory and in at-home environment. More than 560,000 events were analyzed, including a variety of recording devices and different environments. The system's overall accuracy rate is 88.8%, with accuracy rate of 91.2% and 83.6% in in-laboratory and at-home environments respectively, when classifying between inspiratory, expiratory, and noise classes. Here, we provide evidence that breathing-sounds can be reliably detected using non-contact audio technology in at-home environment. The proposed approach may improve our understanding of respiratory activity during sleep. This in return, will improve early SDB diagnosis and treatment.

Automatic Non-Invasive Isolation of Respiratory Cycles

“…The algorithm achieved accuracy either superior [12], or comparable [9], [11] to prior work. A notable problem with comparison to prior work is that there exists no standardised method of evaluating RCI algorithms, and thus different works approach the evaluation differently, making comparison to that work difficult, if not at times impossible [10], [13], [14].…”

“…They reported a 96% precision in detecting breath cycles for participants during rest. Rosenwein et al introduced a breath detection algorithm based on a random forest approach [12]. They derived 351 features from audio recordings and trained the model to detect inspirations and exhalations, reporting an 87% and 76% accuracy in predicting inspiration and expiration events, respectively.…”

Breeze

“…In prior work, MFCC and SVM were used to detect respiratory events versus noise sound and achieved the highest accuracy [57]. The Random Forest (RF) method is used to detect breathing phases and noise [61]. However, the referenced work utilizes features that include the duration of the breathing phases, which cannot be applied in our real-time application.…”

“…Then, for each decision tree in the RF classifier, a subset of features is selected, and the Gini index is used as a cost function to evaluate the split of decisions. The initial values of 13 features as subset per tree and 400 total trees are selected based on [61]. To explore the best possible outcome, we implemented a Gaussian Mixture Model (GMM) with Gammatone Frequency Cepstral Coefficients (GFCC), which is inspired by BreathPrint [14].…”

Pams

“…In general, these works extract a set of features from windows of audio, such as mel-frequency cepstral coefficients (MFCCs), empirical mode decomposition features (EMD), and autocorrelation. These features are then passed to an acoustic event classifier, such as a k-nearest neighbors classifier (KNN), that is trained to determine if snoring or breathing sounds are present [6,[24][25][26][27]. [12] takes a similar approach in using audio to determine if the patient suffers from sleep apnea.…”