A Pilot Study of Deep Learning Models for Camera based Hand Hygiene Monitoring in ICU

Hand hygiene is key to preventing cross-infections in the Intensive Care Unit (ICU). Monitoring of hand washing activities can effectively increase the compliance of clinicians to hand hygiene. In this paper, we explored the feasibility of recognizing clinicians' hand-washing activities using a clinical dataset recorded in ICU using CCTV cameras. We benchmarked three types of hand hygiene detection methods on the dataset with the aim of identifying the 7-step hand washing procedure defined by WHO. Experimental results show that our approach achieves 97% average accuracy for personalized and 67% for generalized modeling. Preliminary results indicate that hand hygiene recognition is subject-dependent, and thus cross-subject modeling or subject-adaptive learning should be used to enhance generalization. The feasibility and challenges of CCTV-camera-based hand hygiene recognition are discussed. The results may contribute to design a hand hygiene scoring and alert system as part of the IoT system in hospitals. The hospital data and code are available at https://github.com/SunnySideUp11/ICU-MH-20.