Motion Estimation and Characterization in Premature Newborns Using Long Duration Video Recordings

Objectives: In the context of neonatal non invasive monitoring, this paper proposes the estimation and characterization of the motion of premature newborns from long duration video recordings. Material and Methods: A set of 13 videos from 9 different patients, corresponding to 190 hours of recordings, have been studied. An algorithm based on the analysis of changes in the image border has been used to remove intervals artifacted by adults' presence. Then, some features were computed to characterize the baby's motion. The approach was applied to compare two groups of premature newborns, with different, severities of prematurity, recorded at the same postmenstrual age. Results: Detection of adults' presence was achieved with 96.8% of sensitivity. All features were found statistically significant to differentiate the two groups. Conclusion: This study shows that the automated video monitoring on long periods is achievable and provides relevant information about the premature newborns motion activity. (C) 2017 AGBM. Published by Elsevier Masson SAS. All rights reserved.