A preregistered, Open Pipeline for Early Cerebral Palsy Risk Assessment from Infant Videos

Comprehensive Summary

Cerebral palsy (CP) is one of the most common causes of childhood motor disability, affecting roughly 1 in every 500 children. Early identification is critical, yet diagnosis often occurs after the optimal window for early intervention has passed. In this study, researchers developed an artificial intelligence–based pipeline to assess CP risk using short infant videos. The model analyzed movement patterns from 1,053 infants using pose-estimation technology and machine-learning algorithms. After extracting 38 movement features related to posture, velocity, symmetry, and movement complexity, the system was trained to predict General Movements Assessment (GMA) scores, a key early indicator of CP risk. When evaluated on a strictly held “lock-box” test set, the model achieved a moderate predictive performance with a ROC-AUC of 0.77 and a precision-recall AUC of 0.41, despite the low prevalence of high-risk cases (only 10–12%). These results demonstrate that AI can identify meaningful movement patterns from simple handheld video recordings without the need for specialized equipment or dataset-specific tuning.

Outcomes and Implications

The implications of this research are significant for the future of medicine, especially in early screening and health equity. By using accessible technology such as smartphone videos and automated machine learning, this approach has the potential to expand early CP screening beyond specialized clinical centers. This could be especially impactful in low-resource or rural settings where trained specialists are limited. Also, the study emphasizes transparency through open-source code and de-identified data sharing, encouraging collaboration across hospitals and research institutions. While the model is not intended to replace clinicians, it highlights how AI can function as a supportive prescreening tool—helping providers identify at-risk infants earlier, allocate medical resources more efficiently, and ultimately improve long-term developmental outcomes through earlier intervention.