Vision-Based Heart and Respiratory Rate Monitoring During Sleep – A Validation Study for the Population at Risk of Sleep Apnea

Kaiyin Zhu, Michael H. Li, Sina Akbarian, Maziar Hafezi, Azadeh Yadollahi, Babak TaatiIEEE Journal of Translational Engineering in Health and Medicine 2019 · DOI

TL;DR

This paper presents a vision-based system using infrared cameras to non-intrusively monitor heart and respiratory rates during sleep. The method shows promising accuracy for respiratory rate estimation, even with variations in sleep position and blanket coverage, suggesting its potential for at-home sleep apnea monitoring. While heart rate estimation is less accurate, the approach offers a valuable step towards convenient, sensor-free sleep health assessment.

What Problem Does This Paper Address?

Sleep apnea is a prevalent and serious condition often underdiagnosed, impacting sleep quality and leading to various health issues. Current gold-standard sleep monitoring, polysomnography (PSG), is expensive, inconvenient, and requires multiple sensors, limiting its widespread use for long-term or at-home monitoring. This creates a need for accurate, accessible, and non-intrusive methods to track vital signs like heart and respiratory rates during sleep, which are crucial for diagnosing and managing sleep disorders. Existing non-contact methods often struggle with accuracy, especially during sleep when movement and varying body positions are common.

Methods

The core of this study is an algorithm that leverages motion analysis from infrared video recordings. The system works by:

  • Feature Point Tracking: Identifying and tracking numerous feature points (e.g., corners, textured areas) on the sleeping participant's body. These points are tracked over 30-second windows.
  • Motion Filtering: To handle artifacts from large movements, the algorithm discards segments with significant motion and dynamically re-samples feature points when body position changes. It also excludes segments where the participant is not visible or where gold-standard readings fall outside a broad physiological range.
  • Respiratory Rate Estimation: Principal Component Analysis (PCA) is applied to the tracked motion data to extract the respiratory rate.
  • Heart Rate Estimation: Independent Component Analysis (ICA) is used for heart rate estimation. To improve accuracy and overcome the challenge of respiratory harmonics interfering with heart rate signals, the algorithm scores candidate frequencies based on harmonic periodicity, proximity to previous estimates, and likelihood of being a heart rate.

Datasets

The study was conducted on 50 adult participants referred for overnight diagnostic sleep studies at the Toronto Rehabilitation Institute. These participants underwent simultaneous gold-standard polysomnography (PSG) for comparison. The dataset includes a diverse range of sleep apnea severities (from no apnea to severe apnea) and various sleep stages and events.

Key Insights and Ideas

  • Robustness to Sleep Variations: The vision-based method demonstrates remarkable resilience to common sleep-related challenges like changes in body position (supine, lateral, prone) and varying degrees of blanket occlusion. This is a significant advantage for real-world applications.
  • Respiratory Rate Accuracy: The algorithm achieves high accuracy in estimating respiratory rate, outperforming previous non-contact methods and a simple "mean hold" baseline. This suggests that vision-based motion analysis is a viable approach for non-contact respiratory monitoring.
  • Heart Rate Challenges: While progress has been made, heart rate estimation remains less reliable than respiratory rate estimation. The paper highlights that residual respiratory motion harmonics are a primary source of error, particularly during obstructive events where breathing is more erratic.
  • Potential for Home Monitoring: The non-intrusive nature and demonstrated robustness make this technology a strong candidate for future at-home sleep monitoring systems, potentially improving early detection and management of sleep disorders.

Key Results

  • Respiratory Rate:
    • 89.89% ± 10.95% of overnight estimations were within 1 breath per minute of the gold standard.
  • Average Root Mean Square Error (RMSE) was 2.10 ± 1.64 breaths per minute.
  • RMSE was significantly higher during awake periods and obstructive events compared to normal breathing sleep.
  • Heart Rate:
    • 77.97% ± 18.91% of overnight estimations were within 5 beats per minute of the gold standard.
  • Average RMSE was 7.47 ± 4.79 beats per minute.
  • Heart rate estimation RMSE was lower during central hypopneas (minimal respiratory effort) compared to other sleep segments.
  • No Significant Differences: No significant differences in RMSE were found based on body position, sleep stage, or blanket coverage for either heart or respiratory rate estimation.

Limitations and Future Work

  • Heart Rate Accuracy: The paper acknowledges that heart rate monitoring is less reliable and requires further work. Future research could explore more adaptive bounds for expected heart rates to improve accuracy.
  • Motion Artifacts: The algorithm excludes segments with large body movements, meaning these periods are not analyzed. Future work could aim to better isolate and analyze these segments.
  • Prone Position: Only a small number of participants slept in the prone position, limiting the ability to draw firm conclusions about performance in this posture.
  • Blanket/Sheet Thickness: Thicker blankets or covers could potentially reduce motion capture, impacting accuracy.
  • Baseline Comparisons: The study used a limited number of baseline methods for comparison, chosen based on code availability.
  • Irregular Heart Rhythms: Participants with irregular cardiac activities were excluded from heart rate analysis, which could limit generalizability.

Future work could involve larger sample sizes, more diverse participant groups, and exploring more advanced signal processing techniques to enhance heart rate accuracy and handle more complex motion scenarios. The ultimate goal is to develop a robust, non-contact system for continuous, long-term sleep monitoring in home environments.

Summary updated

Abstract

A reliable, accessible, and non-intrusive method for tracking respiratory and heart rate is important for improving monitoring and detection of sleep apnea. In this study, an algorithm based on motion analysis of infrared video recordings was validated in 50 adults referred for clinical overnight polysomnography (PSG). The algorithm tracks the displacements of selected feature points on each sleeping participant and extracts respiratory rate using principal component analysis and heart rate using independent component analysis. For respiratory rate estimation (mean ± standard deviation), 89.89 % ± 10.95 % of the overnight estimation was accurate within 1 breath per minute compared to the PSG-derived respiratory rate from the respiratory inductive plethysmography signal, with an average root mean square error (RMSE) of 2.10 ± 1.64 breaths per minute. For heart rate estimation, 77.97 % ± 18.91 % of the overnight estimation was within 5 beats per minute of the heart rate derived from the pulse oximetry signal from PSG, with mean RMSE of 7.47 ± 4.79 beats per minute. No significant difference in estimation of RMSE of either signal was found according to differences in body position, sleep stage, or amount of the body covered by blankets. This vision-based method may prove suitable for overnight, non-contact monitoring of respiratory rate. However, at present, heart rate monitoring is less reliable and will require further work to improve accuracy.
BibTeX
@article{zhu2019vision,
  author = {Kaiyin Zhu and Michael H. Li and Sina Akbarian and Maziar Hafezi and Azadeh Yadollahi and Babak Taati},
  title = {{Vision-Based Heart and Respiratory Rate Monitoring During Sleep – A Validation Study for the Population at Risk of Sleep Apnea}},
  journal = {IEEE Journal of Translational Engineering in Health and Medicine},
  year = {2019},
  doi = {10.1109/jtehm.2019.2946147},
}
Figure