Comfortable sleep monitoring: using physiological process interconnectedness during sleep for novel software sensors

TL;DR

This study introduces non-invasive models to estimate key sleep-related physiological signals using data from RIP belts or pneumoflow masks, aiming to improve comfort and scalability in sleep monitoring.

Contribution

The paper presents novel software sensors that estimate heart rate, nasal airflow, and esophageal pressure using non-invasive sleep data.

Findings

Heart rate and nasal pneumoflow models achieved a mean Pearson correlation of 0.60 when fitted individually.

The RIP-belt-based esophageal pressure model had a mean Pearson correlation of 0.65, outperforming the pneumoflow mask-based model (0.52).

The models offer interpretable estimates and suggest potential for scalable, less burdensome sleep monitoring.

Abstract

Monitoring sleep-disordered breathing typically requires many sensors, including pneumoflow masks, measuring nasal and oral airflow, and esophageal pressure catheters. While these tools provide detailed information about airflow, effort, and respiratory mechanics, they can be uncomfortable, invasive, and less feasible for long-term, home-based, or large-scale sleep studies. In contrast, respiratory inductance plethysmography (RIP) belts offer a non-invasive and well-tolerated alternative.

In this study, we introduce four models that estimate key physiological signals from either RIP-belt data or pneumoflow mask data. Specifically, we present a heart rate model based on the RIP-belt signal, a nasal pneumoflow model estimating airflow from the RIP-belt signal, and two esophageal pressure models – one based on the RIP-belt signal, and the other one based on pneumoflow mask data. Data from 55 participants with varying degrees of sleep-disordered breathing were analyzed.

When fitted to each participant individually, the heart rate model as well as the nasal pneumoflow model achieved a mean Pearson correlation of 0.60. The esophageal pressure model, using RIP-belt data, yielded a mean Pearson correlation of 0.65, while the model using pneumoflow mask data yielded a mean Pearson correlation of 0.52.

Although these models do not replace gold-standard instruments, they provide physiologically interpretable estimates from non-invasive inputs and demonstrate potential for scalable, lower-burden sleep monitoring, and highlight the potential of considering physiological interconnectedness to extract desired information. Future work will focus on further validation and clinical diagnostic utility.

Figures