AI Detects Movement Disorders in Newborns via Video Analysis

Researchers have developed an artificial intelligence system that leverages standard video cameras to enable rapid, objective diagnosis of spinal muscular atrophy in newborns, addressing a critical gap in early neurological screening. Published in JAMA Pediatrics in 2024, the study demonstrates how computer vision can quantify subtle motor deficits associated with infantile hypotonia, a hallmark of the genetic neuromuscular disorder. Spinal muscular atrophy causes progressive degeneration of motor neurons, leading to severe muscle weakness and, in its most acute form, life-threatening respiratory failure before age two. While gene therapies such as Zolgensma have dramatically improved prognoses, their efficacy depends entirely on administration before irreversible neurological damage occurs. Conventional clinical assessments of newborn muscle tone remain subjective and frequently delayed, prompting the need for scalable, objective screening tools. The research team, which evaluated twenty-five infants in pediatric intensive care units, implemented a three-stage computer vision pipeline. A standard camera recorded sixty-second video clips of spontaneous infant movements against a plain background. An AlphaPose algorithm reconstructed each infant motion into a digital skeleton comprising twelve anatomical joints and four movement angles. From this skeletal model, the system extracted one hundred and eight biomechanical parameters, including movement amplitude, symmetry, and depth. A supervised XGBoost classifier then differentiated between typical motricity and SMA-related motor impairment. The system achieved ninety-seven percent classification accuracy and demonstrated greater than ninety-seven percent sensitivity, primarily by measuring limitations in the depth-axis movement that clinically manifests as reduced limb excursion. To ensure clinical transparency, the researchers applied SHAP analysis, which isolates the specific features driving each diagnostic decision and translates algorithmic outputs into interpretable metrics for practitioners. Although France integrated routine SMA genetic screening into its national neonatal program in 2025, the AI camera system remains highly relevant for regions lacking widespread testing infrastructure or requiring immediate triage between genetic results. The technology does not replace clinician judgment but serves as a rapid, equipment-light adjunct that converts subjective visual observations into quantifiable data. Future iterations will expand beyond SMA to screen for other congenital conditions presenting with infantile hypotonia. By compressing the diagnostic timeline, the system aims to maximize the therapeutic window for time-sensitive neonatal interventions.