Objective: To evaluate whether markerless video analysis can extract quantitative motor features that reflect dystonia severity in X-linked dystonia parkinsonism (XDP).

Background: XDP is an ultra-rare combined movement disorder characterized by coexisting dystonic and parkinsonian features, which make clinical assessment challenging and require specialized expertise. 3D markerless pose estimation is an appealing, accessible motion analysis technique, enabling objective extraction of motor features from video. However, there is no data in assessing generalized dystonia, and its ability to capture dystonia in overlapping phenotypes, such as XDP, remains unclear.

Method: We analyzed monocular video recordings from 44 XDP, 15 DYT-TOR1A patients, and 19 controls, performing a standardized neurological examination, including lower limb tasks and gait. A fine-tuned SMPLest model was used to perform markerless pose estimation and reconstruct body kinematics. Joint trajectories were extracted to derive biomechanical features. Clinical dystonia severity was assessed using the Burke-Fahn-Marsden Dystonia Rating Scale (BFM) leg subscores, while MDS-UPDRS lower limb and gait items served as secondary outcomes. Machine learning models were trained to estimate clinical severity scores.

Results: Video-derived features captured clinically meaningful variation in lower limb motor patterns across tasks. Dynamic tasks such as toe tapping and leg agility showed separation between participants with different levels of dystonia severity. XGBoost models estimated dystonia severity expressed as BFM leg subscores (balanced accuracy = 68.03%), indicating that video-derived kinematic features capture motor alterations associated with lower limb dystonia. Models also estimated MDS-UPDRS lower limb and gait items (average balanced accuracy across tasks =69.81%). The differential association of features with BFM and MDS-UPDRS outcomes suggests that the extracted kinematic markers capture distinct aspects of dystonic and parkinsonian motor presentations.

Conclusion: 3D markerless video analysis can capture kinematic features linked to lower limb dystonia severity in XDP. Despite overlapping dystonic and parkinsonian signs, video-derived biomechanical markers may help disentangle dystonia-specific movement patterns. These results support video analysis as a scalable, rater-independent tool for objective dystonia assessment.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/3d-markerless-video-based-assessment-of-lower-extremity-dystonia-in-x-linked-dystonia-parkinsonism-and-dyt-tor1a/