Artificial intelligence in Prechtl's General Movements Assessment: A systematic review and meta-analysis.
Zhanna Zhussupova, Amin Tamadon, Natalya Chagay et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis assessed AI-assisted Prechtl General Movements Assessment (GMA) for predicting later cerebral palsy (CP) and classifying expert-rated GMA labels. Of 105 studies in the qualitative synthesis, 28 contributed to quantitative analyses. Pooled diagnostic accuracy for CP outcomes was 0.884 (95% CI: 0.838-0.918), while pooled classification accuracy for expert-rated labels was 0.848 (95% CI: 0.761-0.908). The authors reported substantial heterogeneity and very low certainty because of heterogeneity and risk of bias, limiting confidence in the pooled estimates.
Key findings
- Across 17 CP diagnosis outcome studies, pooled diagnostic accuracy was 0.884 (95% CI: 0.838-0.918).
- Across 11 expert GMA label outcome studies, pooled classification accuracy was 0.848 (95% CI: 0.761-0.908).
- Heterogeneity was substantial, and the authors judged certainty very low because of heterogeneity and risk of bias.
- No GM developmental phase or sensor modality showed clear superiority.
Why it matters
AI-assisted GMA may help make movement assessment more objective and scalable, including where access to specialist expertise is limited. However, substantial variation between studies and risk of bias mean the pooled accuracy estimates should not be treated as established performance in routine settings. Standardized protocols, high-quality datasets, and transparent validation are important for assessing its clinical potential.
What to keep in mind
The review reports substantial heterogeneity and risk of bias, resulting in very low certainty; only 28 of 105 studies were included in quantitative synthesis.