AGMA-PESS: a deep learning-based infant pose estimator and sequence selector software for general movement assessment.

Abstract

The General Movement Assessment (GMA) is a validated evaluation of brain maturation essential to shaping early individual developmental trajectories of preterm infants. To ensure a reliable GMA, preterm infants should be recorded for 30 to 60 min before manually selecting at least three sequences with general movements. This time-consuming task of manually selecting short video sequences from lengthy recordings impedes its implementation within the Neonatal Unit. Moreover, an accurate pose estimation tool for preterm infants is paramount to developing the field of GMA automation. We introduce the AGMA Pose Estimator and Sequence Selector (AGMA-PESS) software, based on the state-of-the-art deep learning infant pose estimation network, to automatically select the video sequences for GMA at preterm and writhing ages and estimate the pose of infants in 2D. Its simplicity and efficiency make AGMA-PESS a valuable tool to promote GMA use within the Neonatal Unit, both for clinical practice and research purposes.