Expert detection of fidgety movements from skeleton videos: Exploring new avenues for scalable, privacy-preserving general movements assessment.

Empirically testing innovations that initially appear unworkable can open opportunities for human-AI collaboration in medicine. We asked participants with varying general movement assessment (GMA) expertise to detect fidgety movements from skeleton videos, a seemingly implausible task. GMA tutors achieved high accuracy, whereas others (certified non-tutors and individuals with no GMA qualifications) performed heterogeneously, though performance improved overall with exposure in the full sample. Our findings suggest an unexpected privacy-preserving solution for scalable data sharing to enhance human and AI-assisted clinical judgement.