DOI: 10.1177/30682576261474152 ISSN: 3068-2576

Proof-of-principle evaluation of a real-time 3D video respiratory monitoring device

Kévin Albert, Florian Chavernac, Hoang Vu Huy, Srinivasan Ramachandran, Rita Noumeir, David Brossier, Philippe Jouvet
Show PDF Cite

Background

Continuous monitoring of breathing frequency ( f ), tidal volume ( V T ), and flow parameters is essential in intensive care units (ICUs) for detecting early respiratory deterioration. In spontaneously breathing ICU patients, these parameters are difficult to measure continuously because spirometry requires cooperation and interrupts care. Non-contact depth-based sensing may offer a practical and real-time alternative for bedside respiratory monitoring.

Methods

This prospective study evaluated the accuracy of a Kinect Azure–based RGB-D camera system (3DRespiView) designed for automated and non-invasive respiratory monitoring in an ICU-oriented configuration. Healthy adult volunteers underwent simultaneous recordings with a spirometer (reference) and the 3DRespiView system. Each participant completed three 30-second sessions of spontaneous breathing in the supine position. The system automatically detected the thorax, quantified 3D surface displacement, and extracted respiratory parameters including V T , f , inspiratory time ( T i ), expiratory time ( T e ), and peak inspiratory and expiratory flows (PIF/PEF). Agreement was assessed using Bland–Altman analysis, correlation coefficients, mean absolute error (MAE), and root mean square error (RMSE).

Results

Eight participants completed 24 paired recordings. 3DRespiView demonstrated high accuracy across all parameters. For V T , the Bland–Altman bias was 24.1 mL with limits of agreement −17.7 to +66.0 mL; MAE was 26 mL (3.5%) and RMSE 33 mL (4.9%), with a strong correlation (R 2 = 0.99). Breathing frequency ( f ) showed perfect agreement with spirometry (MAE = 0%, R 2 = 1), and T i and T e showed MAEs of 0.10 s (5.2%) and 0.09 s (4.1%), respectively. PIF and PEF showed MAEs of 21 mL/s (3.6%) and 28 mL/s (5.6%), with correlations ranging from R 2 = 0.93 to 0.98. Volume–time waveform comparison showed close temporal alignment between modalities.

Conclusions

3DRespiView demonstrates the potential for accurate, automated, and real-time estimation of respiratory parameters using a non-contact 3D camera. These findings support the technical validity of the approach and highlight its future potential for continuous bedside respiratory monitoring in the ICU.

More from our Archive