Early movement patterns of preterm infants with cerebral palsy (CP)

Background: Premature infants are at risk of developing adverse neurological outcomes, including cerebral palsy (CP). Early detection facilitates timely intervention, optimizing neuroplasticity and function. The General Movement Assessment (GMA) offers a cost-effective means of identifying cerebral palsy by observing infants’ general movements abnormality. Objectives: In this project, we utilized GMA data (68 observations for CP and 92 observations for controls) to classify significant general movements and group infants potentially affected by cerebral palsy at an early stage. Methods: We evaluated the effectiveness of general movements in predicting cerebral palsy by metrics such as accuracy and Cohen’s kappa. Supervised and unsupervised machine learning models were employed for prediction. Results: Fidgety Movements in Typical movement exhibit the highest accuracy at 84% (95% CI: [78%, 89%]), as well as the highest kappa value at 0.67 (95% CI: [0.55, 0.79]) among all variables, indicating excellent capability in distinguishing CP from normal infants. In supervised learning, a logistic model identified an association between general movements and a diagnosis of CP. The model achieved an overall accuracy of 93% (95% CI: [88%, 97%]) and an overall cross-validated accuracy of 90% (95% CI: [66%, 98%]), with a sensitivity of 86% and specificity of 93%. In unsupervised learning, K-prototypes clustering naturally grouped all observations into three clusters with accuracy 87%. One cluster was made up of 85% controls and the other two clusters consisted of 96% and 84% CP. Conclusions: The GMA can predict cerebral palsy with high accuracy, and significant movements correlated with a diagnosis of cerebral palsy, such as Fidgety movements, Kicking, Asymmetry of Finger Posture, were identified.