Artificial intelligence in Prechtl's General Movements Assessment: A systematic review and meta-analysis

ObjectiveTo evaluate AI-assisted GMA performance for (i) prediction of later cerebral palsy (CP) diagnosis and (ii) classification of expert-rated GMA labels, and to assess heterogeneity and risk of bias.MethodsA systematic review and meta-analysis was conducted in accordance with PRISMA 2020 guidelines. A total of 105 studies were included in qualitative synthesis, of which 28 were eligible for quantitative synthesis. Random-effects meta-analysis of proportions with logit transformation was used to estimate pooled diagnostic accuracy.ResultsOf 105 eligible studies in qualitative synthesis, 28 were included in quantitative synthesis (17 CP diagnosis outcomes; 11 expert GMA label outcomes). For CP diagnosis outcomes, the pooled diagnostic accuracy was 0.884 (95% CI: 0.838-0.918). For expert-rated GMA label outcomes, the pooled classification accuracy was 0.848 (95% CI: 0.761-0.908). Heterogeneity was substantial across analyses.InterpretationAI-assisted GMA shows high pooled performance for both CP diagnosis prediction and expert-label classification; however, certainty remains very low due to heterogeneity and risk of bias. No single GM developmental phase, or sensor modality, demonstrated clear superiority, underscoring the importance of standardized protocols, high-quality datasets, and transparent validation. These findings support the clinical potential of AI-enabled GMA as an objective and scalable screening tool, particularly in settings with limited access to specialized expertise.