Touchless monitoring of neonatal activity: a multi-center study.

Abstract

Neonatal activity level is an important physiological parameter linked to lethargic response to over sedation, agitation associated with under-titration of pain medication or seizures and serves as an early indicator of disease onset. Activity was monitored continuously using a non-contact ("touchless") technology based on a depth sensing camera in 61 neonates from two sites (N = 32 'LA' / 29 'Utah'). Gestational age, mean (SD): 34.1 (3.3) weeks Utah, 34.7 (3.8) weeks LA. Time after birth 2.5 (2.5) weeks Utah, 1.6, (1.9) weeks LA. This was compared to manual observation of motion using two main random forest machine learning analyses: [1] combined data set leave-one-out cross-validation (LOOCV) on a per-neonate basis and [2] inter-site analysis. Combined analysis: mean [CI] sensitivity, specificity, and corresponding area under ROC curve for the test sets are 93.8% [92.3, 95.3], 92.2% [90.0, 94.3], and 98.4% [97.8, 99.0], respectively. Inter-site analysis: training using LA data and testing on Utah data resulted in corresponding results of 94.2% [92.0, 96.5], 81.5% [76.2, 86.7], 97.6% [96.5, 98.7], respectively. Utah training and LA testing produced 95.1% [93.3, 96.9], 91.9%, [88.9, 94.9], 98.9% [98.5, 99.4]. Touchless monitoring can provide the basis for quantitative, continuous monitoring of neonatal activity. We investigated the feasibility of non-contact ('touchless') monitoring of neonatal activity based on depth sensing technology. Touchless monitoring offers a sensor and wire-free solution for monitoring activity. The findings support the need for further investigation of the method to determine associations with the measured activity and clinical outcomes.