Estimation of Respiratory Rate from a Corrupted PPG Signal using Time-Frequency Spectrogram

• Published in: International Conference on Signal Processing and Integrated Networks (Online) pp. 259 - 264
• Main Authors: Pankaj, Kumar, Ashish, Komaragiri, Rama, Kumar, Manjeet
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.03.2023
• Subjects: Computational complexity; Estimation; Hospitals; Morlet wavelet; Photoplethysmography (PPG); Pulse oximeter; Real-time systems; Respiratory rate (RR); Respiratory-induced intensity variation; Signal processing algorithms; Transforms; Wearable computers; Wearable device
• ISSN: 2688-769X
• DOI: 10.1109/SPIN57001.2023.10117416

Illnesses like pneumonia, influenza, lung cancer, chronic obstructive pulmonary disease, asthma, and many heart problems, including congestive heart failure, are life-threatening. Respiration rate (RR) is a vital indicator for checking the health status of a subject and can be a marker of early clinical diagnosis. Continuous monitoring of RR can provide an early indication of abnormalities, thereby saving lives. The real-time RR monitoring device is only available in hospitals due to the size and cost of the device.Analyzing the PPG signal change due to respiratory modulations helps estimate RR. This paper describes an algorithm for RR estimation using a PPG sensor signal. During daily activities, inaccuracy in estimating RR using the PPG signal occurs due to motion artifacts. Thus, to improve the accuracy of the RR estimation in real-time, this work investigated the potential of superlet transform-based separation of true RR signal components and the motion artifacts components, thus improving the accuracy of the RR estimation in real-time. The algorithm comprises signal segmentation, pre-processing, and signal transformation using superlet transform to achieve the least mean absolute error (MAE) and robustness to noise. The algorithm is based on the superlet transform technique that includes separating RR peak and motion artifacts peak and detecting high-intensity peak frequency to estimate RR. The proposed algorithm obtained an MAE, and the root mean square error (RMSE) of 0.67 breaths per minute (BPM) and 1.56 BPM, respectively, indicates that embedding the proposed algorithm in a wearable device can be used reliably at home and in hospitals.