Cross‑cohort transfer of video‑based hyperkinetic movement disorder phenotyping
A proof‑of‑concept study developed a video‑analysis framework that combines markerless pose estimation, kinematic descriptors, and a pretrained foundation model to detect eight hyperkinetic movement disorder phenomenologies. The model was trained on 21 adults with confirmed disorders and 4 healthy controls, then externally validated on 12 pediatric patients with monogenic combined disorders; after lightweight calibration of the decision layer, Hamming accuracy rose from 0.804 to 0.839 and the Jaccard index from 0.548 to 0.633 on a held‑out subset of 7 children. The results suggest that a reus…
A proof‑of‑concept study developed a video‑analysis framework that combines markerless pose estimation, kinematic descriptors, and a pretrained foundation model to detect eight hyperkinetic movement disorder phenomenologies. The model was trained on 21 adults with confirmed disorders and 4 healthy controls, then externally validated on 12 pediatric patients with monogenic combined disorders; after lightweight calibration of the decision layer, Hamming accuracy rose from 0.804 to 0.839 and the Jaccard index from 0.548 to 0.633 on a held‑out subset of 7 children. The results suggest that a reus…
The study demonstrates a scalable, AI‑driven approach to routine video‑based phenotyping of movement disorders that could support Swiss healthcare’s AI initiatives in precision diagnostics and personalized treatment planning.
Original source record. Open the original record to verify the underlying announcement.