Non-Contact Automatic Respiratory Rate Monitoring for Newborns Using Digital Camera Technology and Deep Learning

In healthcare settings, particularly for newborns in the Neonatal Intensive Care Unit (NICU), continuous and accurate monitoring of respiratory rates (RR) is crucial, yet current methods often involve manual region selection, which limits their effectiveness. To address this challenge, we developed a novel approach using a low-cost digital camera for non-contact RR monitoring of ten newborns. This study leverages deep learning for automated region of interest (ROI) selection in the face and chest areas, coupled with signal decomposition techniques to reduce noise artifacts. A graphical user interface (GUI) system was also introduced to facilitate real-time RR monitoring and visualization of results. The experimental outcomes demonstrate the system’s high efficacy, with the object detector achieving precision rates of 98.24% for the face region and 96.61% for the chest region. Additionally, the system recorded a low average mean absolute error (MAE) of 1.11 breaths per minute for the face and 1.03 breaths per minute for the chest, corroborated by reference ECG monitor readings. This method offers a cost-effective, non-contact, and deployable solution suitable for both clinical and home health monitoring applications.