Simultaneous hyperkinetic movement disorders phenotyping: a cross-cohort pediatric transfer study using routine videos, markerless pose estimation and a tabular foundation model.

Abstract

Background: Accurate recognition of movement disorders (MDs) phenomenology remains a demanding task in pediatric practice, compounded by mixed and evolving motor presentations, inter-examiner expertise variability, and the scarcity of structured digital tools validated across age groups. A child with combined MDs may receive different phenomenological diagnoses from different examiners, with direct consequences for etiological guidance, treatment planning, neuromodulation candidacy, and longitudinal monitoring. Video-based examination is already central to expert practice, but its systematic analysis remains frequently inaccessible because of limited human resources.

Objective: To develop and externally test a video-based framework for simultaneous detection of hyperkinetic MDs phenomenologies: dystonia, tremor, myoclonus, chorea, athetosis, ballismus, stereotypies, and tics using routine clinical recordings, with external, cross-cohort transfer from adult to pediatric populations.

Methods: In this proof-of-concept study, the framework combines markerless pose estimation, kinematic descriptors, and a pretrained foundation model. A shared predictive backbone was developed on 21 adults with confirmed hyperkinetic MDs and 4 healthy controls. External validation was performed on an independent cohort: a real-world pediatric sample (n = 12, monogenic combined MDs). For the external dataset, the backbone was deployed without retraining; lightweight calibration adjusted only the final subject-level decision step using a clinician-labeled small representative subset of patients.

Results: After local calibration of the decision layer on the clinician-selected subset, performance improved on the held-out pediatric patients (n = 7): Hamming accuracy rose from 0.804 to 0.839 and the Jaccard index from 0.548 to 0.633 (true positives 8 to 13, false negatives 9 to 4, false positives 2 to 5, across 56 patient-phenomenology decisions). Given the small held-out sample, these estimates should be regarded as exploratory. This calibrated performance was preserved, and the Jaccard index further improved, when the evaluation was restricted to the phenomenologies with more definite clinician agreement (Hamming accuracy 0.9, Jaccard index 0.786), indicating that the gains did not rest on the least-reliable labels.

Conclusion: A two-stage strategy, a reusable predictive backbone combined with lightweight, site-specific adaptation, may offer a practical and scalable route to routine-video-based digital phenotyping of MDs across age groups. A shared backbone trained on standardized adult recordings transferred phenotyping signal to an independent, real-world pediatric cohort without retraining. Local calibration of the decision layer improved performance in this small exploratory cohort, while clinical-grade reliability across the full phenotypic spectrum remains to be confirmed in larger cohorts.