A Study of At-term and Preterm Infants' Motion Based on Markerless Video Analysis

Preterm birth is sometimes associated with neurological disorders caused by lesions of the developing brain. A diagnosis in the first weeks of child's life is important to plan timely and appropriate rehabilitative interventions for infants at risk of neuro-motor disabilities. A largely adopted method for the early diagnosis of neuro-motor disorders is the General Movements assessment, based on the evaluation of infants' spontaneous motor patterns. However, an accurate clinical assessment of infant motion requires highly specialized personnel, not always available at all sites. To insure an objective motion analysis, several studies proposed the use of marker-based techniques. Unfortunately, markers are uncomfortable and can affect the naturalness of the motion. Therefore, much effort has been dedicated in developing marker-less techniques targeting unobtrusive and reliable motion analysis. In this work we propose a marker-less video-based methodology to analyze infants' spontaneous movements in RGB videos. First, we detect relevant landmarks on the infants' body. Then, we compute kinematic parameters that describe infants' motion patterns. We validate the effectiveness of the computed parameters on a dataset of 68 infants, 27 of which with a clinically-assessed evidence of neuro-motor disorders: our method successfully discriminates infants with and without motor disorders with an accuracy of 78.2%.