A Video-Based Approach to Characterize Preterm Infants’ Motion as a Sequence of Poses

The analysis of spontaneous movements in preterm infants is crucial for detecting abnormal motion patterns associated with neuromotor deficits. Early diagnosis is essential for planning timely interventions, thereby increasing the chances of recovery. Traditional diagnostic tools often rely on highly specialized clinicians' subjective evaluation, which can be a limitation. In this study, we proposed a novel video-based pipeline to objectively and automatically characterize the spontaneous movements of preterm infants based on 2D video recordings. We built a dictionary of prototypical poses based on a dataset of 142 preterm infants acquired at term of equivalent age, of which 59 were diagnosed with abnormal motion patterns. Then, we computed interpretable metrics highlighting motor differences between infants with and without neuromotor deficits that expert clinicians can leverage during their diagnosis.