Advanced respiratory disease monitoring: a machine learning driven software defined radios approach for accurate respiratory rate

Respiratory disease management has become a critical area of focus, necessitating innovative solutions for continuous and non-invasive monitoring. Human respiratory patterns, which reflect the rate, depth, and rhythm of breathing, are vital indicators of an individual's physical and mental health. Effective monitoring of these patterns can aid in diagnosing related disorders and, significantly improving patient care. This study proposes a contactless respiratory pattern recognition system that utilizes an advanced radio frequency (RF) sensing framework combined with smart software-defined radio (SDR) technology and machine learning algorithms. The system ensures precise measurement and recognition of the respiratory patterns. Initially, data-preprocessing techniques were developed to capture the time-domain respiratory signals from the received data accurately. Additionally, a method based on Fast Fourier Transform (FFT) was employed to detect abnormal respiratory rates. Subsequently, various machine learning algorithms were applied, with the Medium Tree algorithm proving to be highly effective. The experimental results indicate that the system can accurately identify multiple respiratory cycles, including eupnea, tachypnea, and bradypnea, achieving an accuracy of 89.8 % in scenarios involving up to five individuals. The Area Under the Receiver Operating Characteristic (AUROC) scores demonstrated the superior performance of the Medium Tree model (98.5 to 89.8 %) compared to Linear SVM (97.4 to 85.3 %), Linear Discriminant (94.6 to 87.6 %), Naive Bayes (98.3 %), and Coarse KNN (96.4 %), with a paired t-test (p = 0.0238) confirming its significant advantage over the Linear Discriminant model. The system performance was validated in diverse indoor environments, thereby demonstrating its robustness and reliability. The proposed contactless respiratory pattern recognition system demonstrates the feasibility of long-term continuous respiratory monitoring.