Video-based automated assessment of General Movements in preterm infants: an experimental methodological evaluation of different biomarkers from literature review

One of the most promising methods that can be used to anticipate a neurodevelopmental disorder diagnosis is called General Movement Assessment (GMA) and it is exploited to evaluate the complexity and rhythmicity of General Movements (GM), which are motions that infants perform from the last weeks of gestation until the appearance of intentional movements and that differ in strength and speed, accurately reproducing infants’ complex brain development. At the moment, however, GMA examination can be carried out only by expert licensed clinicians and its usage is limited by its high subjective component. In literature, many technology-based solutions, mainly based on a 2D video analysis, have been proposed to overcome this problem, but none of them are actually used in the daily clinical practice as they lack a standard protocol both for video acquisition and data processing. The present thesis aims at assessing, from a methodological point of view, the most significant GM biomarkers proposed in literature, starting from a limited set of videos chosen from the ones collected for a wider study at the Neonatal Unit of Sant’Orsola Hospital in Bologna. In particular, 25 videos, recorded by clinical staff members using a commercial video camera at 240 fps, of preterm newborns were analyzed. Reference points trajectories were extracted using DeepLabCut and then, 31 metrics selected from literature were assessed. Influence of trajectories interpolation and down sampling was evaluated. Results show that both video acquisition settings and pre-processing stages affect the final metrics values and, therefore, these outcomes highlight the necessity of a systematic approach to obtain a good objective system that can be use in daily clinical practice.