Enhanced Markerless Tracking of Infant General Movements in Standard Videos Through Lightning Pose Compared to...

Monitoring abnormal or absent General Movements (GMs) in infants between 6-20 weeks post-term age serves as a reliable prognostic tool for assessing neurodevelopmental quality and determining the risk of neurological impairments, such as cerebral palsy (CP), in early infancy. Early therapeutic interventions during this timeframe are most effective due to the brain's plasticity. Building on our previous work, this paper compares the reliability and applicability of the Lightning Pose (LP) platform to our prior attempts using the DeepLabCut (DLC) environment for the reliable markerless pose-estimation and tracking of infants' anatomical landmarks in clinically recorded videos from handheld devices (e.g., iPads). Here, we specifically compare the capabilities of Resnet-152, trained in both platforms, for identifying a novel clinically useful set of 24 anatomical landmarks across a comprehensive unseen set of video frames, tracking a baby's natural movement during validation recordings. Using our novel unsupervised performance assessment algorithm on over 3200 unseen frames from two full-term male infants, we show that LP outperforms DLC in body parts identification, achieving 99.75% accuracy compared to DLC's 97.80%, despite nearly identical training sets extracted by both algorithms (average cosine similarity index of 0.95).Clinical relevance-The superior performance of LP in this study represents a significant stride toward developing a cost-effective and widely accessible technology for automatically monitoring of infants' GMs and identifying those at elevated risk of developing neurological conditions in early infancy.