Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transfer Learning for Paediatric Sleep Apnoea Detection Using Physiology-Guided Acoustic Models

Chaoyue Niu, Veronica Rowe, Guy J. Brown, Heather Elphick, Heather Kenyon, Lowri Thomas, Sam Johnson, Ning Ma

TL;DR

This work tackles paediatric OSA detection from acoustic data by transferring adult-trained models to paediatric recordings and incorporating SpO2 desaturation as a physiological prior. It presents a CNN-based framework with single- and multi-task heads, comparing freezing versus full encoder fine-tuning and integrating delayed SpO2 labels to better align physiology with acoustics. Across a 5-fold cross-validation on 15 paediatric nights, physiology-informed transfer learning (MTL, full fine-tuning, subject-specific delays) achieved the best AHI estimation (MAE $=2.81$, RMSE $=3.86$), outperforming non-adapted baselines. The findings support the feasibility of at-home, smartphone-based triage for paediatric OSA and highlight the value of physiological priors for data-efficient learning.

Abstract

Paediatric obstructive sleep apnoea (OSA) is clinically significant yet difficult to diagnose, as children poorly tolerate sensor-based polysomnography. Acoustic monitoring provides a non-invasive alternative for home-based OSA screening, but limited paediatric data hinders the development of robust deep learning approaches. This paper proposes a transfer learning framework that adapts acoustic models pretrained on adult sleep data to paediatric OSA detection, incorporating SpO2-based desaturation patterns to enhance model training. Using a large adult sleep dataset (157 nights) and a smaller paediatric dataset (15 nights), we systematically evaluate (i) single- versus multi-task learning, (ii) encoder freezing versus full fine-tuning, and (iii) the impact of delaying SpO2 labels to better align them with the acoustics and capture physiologically meaningful features. Results show that fine-tuning with SpO2 integration consistently improves paediatric OSA detection compared with baseline models without adaptation. These findings demonstrate the feasibility of transfer learning for home-based OSA screening in children and offer its potential clinical value for early diagnosis.

Transfer Learning for Paediatric Sleep Apnoea Detection Using Physiology-Guided Acoustic Models

TL;DR

This work tackles paediatric OSA detection from acoustic data by transferring adult-trained models to paediatric recordings and incorporating SpO2 desaturation as a physiological prior. It presents a CNN-based framework with single- and multi-task heads, comparing freezing versus full encoder fine-tuning and integrating delayed SpO2 labels to better align physiology with acoustics. Across a 5-fold cross-validation on 15 paediatric nights, physiology-informed transfer learning (MTL, full fine-tuning, subject-specific delays) achieved the best AHI estimation (MAE , RMSE ), outperforming non-adapted baselines. The findings support the feasibility of at-home, smartphone-based triage for paediatric OSA and highlight the value of physiological priors for data-efficient learning.

Abstract

Paediatric obstructive sleep apnoea (OSA) is clinically significant yet difficult to diagnose, as children poorly tolerate sensor-based polysomnography. Acoustic monitoring provides a non-invasive alternative for home-based OSA screening, but limited paediatric data hinders the development of robust deep learning approaches. This paper proposes a transfer learning framework that adapts acoustic models pretrained on adult sleep data to paediatric OSA detection, incorporating SpO2-based desaturation patterns to enhance model training. Using a large adult sleep dataset (157 nights) and a smaller paediatric dataset (15 nights), we systematically evaluate (i) single- versus multi-task learning, (ii) encoder freezing versus full fine-tuning, and (iii) the impact of delaying SpO2 labels to better align them with the acoustics and capture physiologically meaningful features. Results show that fine-tuning with SpO2 integration consistently improves paediatric OSA detection compared with baseline models without adaptation. These findings demonstrate the feasibility of transfer learning for home-based OSA screening in children and offer its potential clinical value for early diagnosis.

Paper Structure

This paper contains 12 sections, 5 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Data
  3. Methods
  4. Feature extraction and preprocessing
  5. Problem formulation
  6. Physiological information
  7. Transfer learning strategy
  8. EXPERIMENTAL SETUP
  9. Training setup
  10. Evaluation framework
  11. RESULTS AND DISCUSSION
  12. Conclusion

Figures (3)

  • Figure 1: Pipeline for data collection and model development. Breathing sounds are recorded using a smartphone, optionally supplemented with chest sensor and IMU signals. Clinician-annotated HST provides apnoea–hypopnoea events and AHI scores. Solid lines denote data used in the current model, while dashed lines indicate signals reserved for future integration. The fine-tuning pipeline loads an adult-pretrained multi-task CNN, optionally freezes the shared CNN backbone, and fine-tunes task-specific layers.
  • Figure 2: Time delay of oxygen desaturation and Mel-frequency spectrum. Red dashed rectangles mark the 15-s oxygen desaturation window, which is shifted from 0 to longer time delays to identify the delay that best discriminates between normal and abnormal breathing.
  • Figure 3: Median - 95% CI bar plots showing time spent below SpO2 baseline as a percentage within a 15 s window versus time delay of SpO2 for non-OSA, OSA and hypopnoea segments. Median and 95% CI for non-OSA events are 0.