Multisensor data fusion for enhanced respiratory rate estimation in thermal videos

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Medical Imaging

Background:

• Atypical respiratory rate (RR) is a key indicator of physiological distress.
• Current RR monitoring methods have limitations, often requiring sensor attachment.
• RR is considered a neglected vital parameter due to monitoring challenges.

Purpose of the Study:

• To introduce a novel, non-contact approach for enhanced respiratory rate estimation.
• To utilize multisensor data fusion in thermal videos for improved RR monitoring.
• To validate the proposed method against a gold standard in controlled experiments.

Main Methods:

• Developed a novel approach using multisensor data fusion in thermal videos.
• Incorporated temperature variations around nostrils/mouth and shoulder movements.
• Conducted experiments with healthy candidates simulating normal and varied breathing patterns.
• Validated the algorithm using thoracic effort as the gold standard.

Main Results:

• Achieved excellent agreement between infrared thermography (IRT) and ground truth.
• Phase A (normal breathing): Mean correlation of 0.983, RMSE of 0.240 bpm.
• Phase B (varied breathing): Mean correlation of 0.995, RMSE of 0.890 bpm.
• Demonstrated enhanced RR estimation and robustness through multisensor data fusion.

Conclusions:

• Infrared thermography (IRT) shows potential as a clinical alternative to conventional sensors.
• Multisensor data fusion significantly enhances respiratory rate estimation accuracy and robustness.
• The novel approach offers a promising non-contact solution for physiological distress monitoring.