Automating the General Movements Assessment: Potential, pitfalls, and a path forward
Developmental Medicine & Child Neurology · 17.06.2026
DOI: 10.1111/dmcn.70339 ↗Kurz gefasst
Fachartikel in der Zeitschrift Developmental Medicine & Child Neurology, online erschienen am 17.06.2026. Verfasst von Alicia Spittle und 1 weiteren Autor:innen. Beteiligte Einrichtung: Department of Physiotherapy, Melbourne School of Health Science, University of Melbourne, Melbourne, Australia. Die Arbeit nennt 5 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Health and Medical Research Council.
Abstract (Originalsprache)
The General Movements Assessment (GMA) is highly predictive in early infancy, particularly for cerebral palsy (CP).It enables referral for early intervention and parental support during a critical period of neuroplasticity.Over the past decade, interest in artificial intelligence (AI) to automate the GMA (especially during the fidgety period) has increased alongside advances in computer vision.The goal is scalable early screening across high-and low-resource settings by reducing reliance on certified assessors and labour-intensive manual scoring. 1 Although earlier reviews found no single optimal approach, 2 Kaur et al.'s 3 scoping review favours deep learning over traditional machine-learning methods.Automated GMA should be understood as a sequential workflow, from recording quality and pose-estimation through to preprocessing, feature extraction, and classification, with each stage dependent on the reliability of the previous one.Failure at any point undermines downstream performance.Despite methodological advances, automated GMA remains confined to retrospective, in silico studies.No studies have included clinical trials of any kind; the logical next step is the introduction of silent trials, running in parallel to standard care.This is critical, as performance reported in controlled research settings is unlikely to translate directly into routine clinical practice.Most existing models identified in the review by Kaur et al. 3 are trained and validated on a single dataset collected under specific recording protocols and population contexts.Models reporting the highest accuracy are often trained on small, heterogeneous datasets, with performance declining as dataset size increases.For a tool intended for widespread screening, poor generalizability remains a major barrier to deployment of automated GMA into clinical practice.
Autor:innen und Institutionen
- Department of Physiotherapy, Melbourne School of Health Science, University of Melbourne, Melbourne, Australia.
- Developmental Imaging Group, Murdoch Children's Research Institute, Melbourne, Australia.
ORCID-Profile: Alicia Spittle
Förderung
- National Health and Medical Research Council
Zitieren
APA
Spittle, A., & Passmore, E. (2026). Automating the General Movements Assessment: Potential, pitfalls, and a path forward. Developmental Medicine & Child Neurology. https://doi.org/10.1111/dmcn.70339
Vancouver
Spittle A, Passmore E. Automating the General Movements Assessment: Potential, pitfalls, and a path forward. Developmental Medicine & Child Neurology. 2026. doi:10.1111/dmcn.70339
BibTeX
@article{Spittle2026,
author = {Spittle, Alicia and Passmore, Elyse},
title = {Automating the General Movements Assessment: Potential, pitfalls, and a path forward},
journal = {Developmental Medicine & Child Neurology},
year = {2026},
doi = {10.1111/dmcn.70339}
}
EUVIMED