Video-Based Respiratory Rate Estimation for Infants in the NICU

Objective: Non-contact respiratory rate estimation (RR) is highly desirable for infants because of their sensitive skin. We propose a novel RGB video-based RR estimation method for infants in the neonatal intensive care unit (NICU) that can accurately measure the RR contact-less. Methods and Procedures: We utilize Eulerian video magnification (EVM) method and develop an adaptive peak prominence threshold value estimation method to address challenges of RR estimation (e.g., dark environments, shallow breathing, babies swaddled or under blankets). We recruited 13 infants recorded for 4 consecutive hours per case. We then evaluate the performance of the algorithm for several (i.e., 19 to 25) randomly selected videos, each lasting 1 minute, for each case. Results: Intraclass correlation coefficients of the proposed method over manually and automatically selected ROIs are 0.91 (95%CI: 0.89 - 0.93) and 0.88 (95%CI: 0.85 - 0.9), indicating excellent and good reliability, respectively. The Bland-Altman analysis of the proposed algorithm shows higher agreement between the estimated values via the proposed method and visually counted RR than the agreement between the RR obtained from the impedance sensors and reference RR, and agreement between a former EVM-based method and reference RR values. Conclusion: Our algorithm shows promising results for RR estimation in a real-life NICU environment under various conditions that can confound the estimation. Clinical impact: We present a robust algorithm for non-contact neonatal respiratory rate monitoring, capable of performing well under various environmental lighting conditions in NICU, even when the infant is clothed or covered.