Identification of Movement Patterns Associated with Neurological Disorders Using Machine Vision

The identification and quantification of abnormal movement patterns are central to the diagnosis, monitoring, and treatment of neurological disorders such as Parkinson’s disease (PD), stroke, cerebral palsy (CP), and dystonia. Machine vision the combination of markerless pose estimation, computer vision-based feature extraction, and machine learning has emerged as a scalable, non-invasive, and objective approach for capturing clinically meaningful motor biomarkers from videos and standard cameras. This review synthesizes recent advances (2018–2025) in machine-vision pipelines for detection, classification, and quantification of movement patterns associated with neurological disorders. We first summarize the technical building blocks: markerless pose estimation (e.g., OpenPose, DeepLabCut, MediaPipe), representation and feature extraction methods (kinematic, temporal, spectral), and machine-learning models (classical and deep approaches). Next, we review disorder-specific applications (PD motor signs, post-stroke gait and limb impairment, CP gait analysis, dystonia detection, and facial/ocular biomarkers) and discuss datasets, evaluation metrics, and clinical validation efforts. Finally, we consider major challenges domain shift, data privacy, standardization, interpretability, and ethics and propose future research directions, including multimodal fusion, federated learning, standard benchmarks, and regulatory pathways for clinical translation. Machine vision offers transformative potential for neurology and rehabilitation, but consistent external validation and cross-site standardization remain prerequisites for clinical deployment.