Video-based computational analysis of spontaneous movements in preterm infants: A longitudinal neuromotor assessment

Monitoring the spontaneous movements of preterm infants is crucial for the early detection of potential neuromotor deficits. Recent advances in human pose estimation have made markerless video-based methods a valid, non-intrusive, and cost-effective option for analyzing these movements. This paper aims to explore the efficacy of video-based markerless techniques in assessing infants' spontaneous movements, with a focus on identifying early signs of developmental abnormalities.

We conducted a longitudinal study with two acquisition sessions to evaluate the stability and consistency of our video-based analysis over time. Our approach builds on previous methodologies by incorporating advanced techniques for feature detection, parameter extraction, feature selection, and classification. Emphasis was placed on the interpretability and clinical relevance of the extracted motion parameters.

The results highlight the effectiveness of our approach in identifying subtle changes in infants' motion patterns that may indicate neuromotor deficits. We observed differences in the detection of these deficits across the acquisition sessions, with our method achieving a maximum test accuracy of 90%.

Our findings support the potential of markerless video-based analysis as a valuable tool in the support of the early detection of neuromotor deficits in preterm infants. The high accuracy and clinical relevance of our approach suggest it could play a critical role in early intervention strategies.