Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Estimating Respiratory Effort from Nocturnal Breathing Sounds for Obstructive Sleep Apnoea Screening

Xiaolei Xu, Chaoyue Niu, Guy J. Brown, Hector Romero, Ning Ma

TL;DR

The paper tackles the challenge of scalable, sensor-free obstructive sleep apnea screening by estimating respiratory effort directly from nocturnal audio and fusing it with acoustic features to improve detection. A two-step framework is presented: an audio-to-effort estimator trained with a Concordance Correlation Coefficient objective, followed by freezing its latent embeddings and combining them with audio embeddings for OSA detection and AHI estimation. Evaluated on 157 nights from 103 home-recorded participants, the approach achieves a CCC of $0.478 \pm 0.133$ for effort and yields gains in sensitivity and AUC at low AHI thresholds compared with audio-only baselines, while requiring only smartphone audio at test time. This sensor-free, longitudinal monitoring setup has practical implications for scalable at-home OSA screening and continuous disease burden tracking.

Abstract

Obstructive sleep apnoea (OSA) is a prevalent condition with significant health consequences, yet many patients remain undiagnosed due to the complexity and cost of over-night polysomnography. Acoustic-based screening provides a scalable alternative, yet performance is limited by environmental noise and the lack of physiological context. Respiratory effort is a key signal used in clinical scoring of OSA events, but current approaches require additional contact sensors that reduce scalability and patient comfort. This paper presents the first study to estimate respiratory effort directly from nocturnal audio, enabling physiological context to be recovered from sound alone. We propose a latent-space fusion framework that integrates the estimated effort embeddings with acoustic features for OSA detection. Using a dataset of 157 nights from 103 participants recorded in home environments, our respiratory effort estimator achieves a concordance correlation coefficient of 0.48, capturing meaningful respiratory dynamics. Fusing effort and audio improves sensitivity and AUC over audio-only baselines, especially at low apnoea-hypopnoea index thresholds. The proposed approach requires only smartphone audio at test time, which enables sensor-free, scalable, and longitudinal OSA monitoring.

Estimating Respiratory Effort from Nocturnal Breathing Sounds for Obstructive Sleep Apnoea Screening

TL;DR

The paper tackles the challenge of scalable, sensor-free obstructive sleep apnea screening by estimating respiratory effort directly from nocturnal audio and fusing it with acoustic features to improve detection. A two-step framework is presented: an audio-to-effort estimator trained with a Concordance Correlation Coefficient objective, followed by freezing its latent embeddings and combining them with audio embeddings for OSA detection and AHI estimation. Evaluated on 157 nights from 103 home-recorded participants, the approach achieves a CCC of for effort and yields gains in sensitivity and AUC at low AHI thresholds compared with audio-only baselines, while requiring only smartphone audio at test time. This sensor-free, longitudinal monitoring setup has practical implications for scalable at-home OSA screening and continuous disease burden tracking.

Abstract

Obstructive sleep apnoea (OSA) is a prevalent condition with significant health consequences, yet many patients remain undiagnosed due to the complexity and cost of over-night polysomnography. Acoustic-based screening provides a scalable alternative, yet performance is limited by environmental noise and the lack of physiological context. Respiratory effort is a key signal used in clinical scoring of OSA events, but current approaches require additional contact sensors that reduce scalability and patient comfort. This paper presents the first study to estimate respiratory effort directly from nocturnal audio, enabling physiological context to be recovered from sound alone. We propose a latent-space fusion framework that integrates the estimated effort embeddings with acoustic features for OSA detection. Using a dataset of 157 nights from 103 participants recorded in home environments, our respiratory effort estimator achieves a concordance correlation coefficient of 0.48, capturing meaningful respiratory dynamics. Fusing effort and audio improves sensitivity and AUC over audio-only baselines, especially at low apnoea-hypopnoea index thresholds. The proposed approach requires only smartphone audio at test time, which enables sensor-free, scalable, and longitudinal OSA monitoring.

Paper Structure

This paper contains 13 sections, 4 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Methods
  4. Acoustic-based respiratory effort estimation
  5. SDB events detection and AHI estimation
  6. EXPERIMENTAL SETUP
  7. Baselines
  8. Training setup
  9. Evaluation metrics
  10. RESULTS AND Discussion
  11. Respiratory effort prediction from audio
  12. OSA severity classification
  13. Conclusion

Figures (2)

  • Figure 1: System diagram of the proposed latent fusion approach for acoustic-based sleep apnoea detection. A CNN-LSTM encoder extracts general acoustic embeddings and respiratory-effort-oriented latent representations from nocturnal audio. These are fused and fed into a classifier to predict apnoea/hypopnoea events.
  • Figure 2: Typical examples of predicted respiratory effort from audio (dashed line) compared to measured ground truth (solid line).