Development and Validation of a Multimodal Deep Learning Model Using Peripheral Oxygen Saturation and Ballistocardiography Signals for Predicting Apnea-Hypopnea Index in Obstructive Sleep Apnea

Condition(s) studied

Study summary

Obstructive sleep apnea (OSA) is a common sleep disorder associated with cardiovascular, metabolic, and neurocognitive complications. The standard diagnostic test for OSA is in-laboratory polysomnography (PSG), which can be costly, labor-intensive, and inconvenient for repeated or large-scale screening. Therefore, there is a need for a convenient and non-invasive method to help identify individuals who may be at risk for OSA.

This prospective observational study aims to develop and clinically validate a multimodal deep learning model that integrates peripheral oxygen saturation (SpO₂) signals and ballistocardiography (BCG) data collected using a smart mattress system to estimate the apnea-hypopnea index (AHI). Adult participants undergoing routine overnight PSG for suspected OSA will be enrolled. During sleep, synchronized SpO₂ and BCG signals will be recorded non-invasively.

The performance of the deep learning model will be evaluated by comparing the predicted AHI values with AHI values derived from PSG. This study seeks to evaluate whether this practical, non-invasive, and scalable approach may support early identification of OSA risk and improve access to sleep disorder assessment in both clinical and home-based settings.

Eligibility

Primary outcome measure(s)

• Mean absolute error of apnea-hypopnea index estimation — Two overnight assessments per participant: one in-laboratory polysomnography night and one subsequent home sleep monitoring night.
  Mean absolute error (MAE) will be calculated as the average absolute difference between the apnea-hypopnea index (AHI) predicted by the multimodal deep learning model using peripheral oxygen saturation (SpO2) and ballistocardiography (BCG) signals collected during home sleep monitoring and the AHI measured by in-laboratory polysomnography (PSG). The analysis will include participants with evaluable paired PSG and home monitoring data. MAE will be reported in events per hour, with lower values indicating better predictive performance.

Trial sites (1)

More Fu Jen Catholic University trials in Taiwan

Other trials for the same condition