Image Analysis of the Nasal Region for Respiratory Rate Detection

Respiration health monitoring is vital due to the rising prevalence of an aging population, chronic respiratory diseases, and challenges faced by those in rural and underserved areas. Advancements in telehealth applications have allowed medical professionals to monitor and diagnose patients remotely, offering a more accessible and convenient approach to healthcare. However, one of the main challenges in telehealth is the need for physical examination, falling short in terms of providing real-time physiological data to medical professionals. Hence, this study aims to bridge that gap by developing a vision-based, non-invasive respiratory rate (RR) monitoring system using the MLX90640 thermal camera and providing quantifiable respiratory metrics such as breaths per minute (BPM). Using a thermal camera also minimizes the interference of surroundings, such as brightness, which may hinder analyzing BPM when using a conventional RGB camera. A YOLOv8 model was trained using 1,690 thermal images to classify breathing patterns. The system effectively extracted an oscillating signal, representing the thermal fluctuations of the nasal region during inhalation and exhalation. This is further processed by signal manipulation techniques such as bandpass filtering, Fast Fourier transformation (FFT), and peak detection. Ultimately, the camera system was successfully calibrated to monitor RR, achieving an RMSE score of 1.651 relative to the standard method of manually counting BPM by observing the rise and fall of a patient's chest area.