An orthogonal non-negative matrix factorization approach for respiratory rate estimation using a wireless stethoscope

Respiratory rate (RR) is a sensitive parameter for indicating critical physiological deteriorations, however, its measurement remains the most inaccurate within the field of vital signs monitoring. This paper proposes three main contributions. Firstly, the development of a wireless stethoscope is proposed to optimize the fidelity of recorded respiratory sounds. Secondly, a new labeled biomedical database is presented that has been created capturing respiratory sounds using the proposed stethoscope and a commercial one. This database aims to be a valuable tool to others researchers in evaluating novel approaches for measuring this vital sign. Thirdly, a new method based on Orthogonal Non-negative Matrix Factorization (ONMF) is presented for the purpose of RR estimation highlighting that ONMF has never before been applied in this biomedical task to our knowledge. Furthermore, the proposed ONMF method has been improved by minimizing the detection of individual respiratory stages, inspiration or expiration, instead the complete respiratory cycle. Preliminary results indicated that the quality of the sound captured by the proposed stethoscope is comparable to that of the commercial stethoscope used in this study. The results indicate the superior performance of the ONMF approach over several signal processing-based methods and artificial intelligence-based methods, irrespective of the acoustic environment, achieving at least a 2.5 bpm reduction in Absolute Error (AE) compared to signal processing-based methods and at least a 0.6 bpm reduction compared to artificial intelligence-based methods. Finally, the proposed fine-tuning algorithm demonstrated a significant improvement in RR estimation, underscoring the task of correct detection of periodic respiratory patterns at the cycle rather than respiratory stage level to provide an accurate measurement of human respiration rate. • Development of a prototype of wireless stethoscope to capture respiratory sounds. • A database is created using respiratory sounds labeled with respiratory rates. • The respiratory rate is estimated applying a novel ONMF approach. • The ONMF performance is improved maximizing the detection of the respiratory cycle. • Results indicate a promising detection performance in noiseless and noise scenarios.