This study presents StimVision, a smartphone video-based, markerless kinematic pipeline that quantifies hand movement features to objectively rank and optimize DBS programming within sessions in Parkinson’s patients.

By providing scalable, quantitative motor biomarkers and a patient-specific improvement score, the method can make DBS programming more efficient and reproducible, enable comparative assessment against dopaminergic therapy, and facilitate objective endpoints for clinical optimization and trials.

Deep brain stimulation (DBS) improves motor function in Parkinson's disease, yet programming remains labor-intensive and largely subjective. We evaluated a smartphone video-based kinematic framework (StimVision) for objective, within-session optimization of DBS settings and characterization of therapeutic motor signatures. Fifteen patients with subthalamic DBS performed repetitive hand opening-closing while multiple stimulation programs were tested in the medication-off state. Markerless pose estimation from 60 fps smartphone video generated 23 quantitative kinematic features. A patient-specific Dynamically Weighted Improvement Score (DWIS) ranked programs by composite improvement relative to DBS-off. The framework identified a unique optimal program for each patient, with robust ranking stability. Group-level improvements at the optimal setting were dominated by gains in speed and rhythm metrics, including mean velocity, closing speed, and movement frequency, alongside reduced intra-sequence decay. Sparse principal component analysis revealed three kinematic domains-Movement Speed, Movement Consistency, and Rhythm & Timing. Structural comparison with a levodopa cohort demonstrated substantial overlap in speed and consistency domains but divergence in timing-related features. Smartphone-based kinematics enable objective DBS optimization and provide a shared quantitative framework for comparing electrical and pharmacological therapies.