Self-Adaptive Illumination Enhancement for Neonatal Remote Physiological Measurement in NICUs

In the NICU, continual heart rate monitoring is crucial for assessing infant health, but conventional methods like electrode attachment can cause discomfort or harm to a newborn’s delicate skin due to skin contact. Exploring non-contact methods for heart rate monitoring in NICUs is crucial, but dim lighting in NICUs poses a challenge for using the remote photoplethysmography (rPPG) method. This study proposes an end-to-end adaptive lighting condition enhancement method for neonatal rPPG extraction, enabling accurate heart rate assessment in NICUs even under poor lighting conditions, thereby increasing the precision and robustness of rPPG signal extraction. Camera-based physiological monitoring for adults has made significant advancements in recent years. However, public neonatal databases are rare, and most existing methods are tailored for adults. Neonatal databases typically use well-lit conditions to ensure clarity and effective model training. This paper introduces the NICU-NPS database, containing 19.9 hours of video recordings from 25 infants under varying lighting conditions, totaling 2,420 videos or 283 GB. Experimental results show that our proposed rPPG extraction method performs well in terms of accuracy and stability under the real NICU lighting mode, reflecting clinical and practical application.