AI-based video kinematic analysis of 53 PD patients undergoing levodopa challenge and STN-DBS showed that levodopa and DBS produce distinct, domain-specific improvements in bradykinesia (speed, amplitude, variability), with AI metrics detecting treatment effects not evident on conventional…

Provides a translational digital biomarker approach that sensitively distinguishes medication versus DBS effects, supporting personalized treatment profiling, improved monitoring, and more sensitive endpoints for clinical trials in Parkinson's disease.

BACKGROUND: Bradykinesia in Parkinson's disease (PD) involves reduced movement speed, amplitude, and rhythmicity. While the MDS-UPDRS Part III is the standard clinical tool for motor assessment, it has limited sensitivity to specific kinematic features. Levodopa and subthalamic nucleus deep brain stimulation (STN-DBS) are common treatments for PD, yet their differential effects across motor domains are not fully characterized. This study applies AI-based video analysis to evaluate the effects of levodopa and STN-DBS on limb bradykinesia. METHODS: This retrospective study assessed fifty-three patients with Parkinson's disease undergoing STN-DBS. Motor performance was video-recorded during Levodopa-off and Levodopa-on states (levodopa challenge test performed prior to surgery), as well as after DBS activation (OFFMED/OFFSTIM, OFFMED/ONSTIM, ONMED/ONSTIM). Both clinical assessments and subsequent video-based analyses focused on the MDS-Unified Parkinson's Disease Rating Scale (MDS-UPDRS), Part III, specifically evaluating items 3.4 Finger Tapping, 3.5 Fist-clenching test, 3.7 Toe Tapping, and 3.8 Leg Agility. Motor function was first evaluated using conventional UPDRS-III item scores rated by two experienced specialists, with the primary clinical comparison defined between the levodopa-on and OFFMED/ONSTIM states, to explore the differential therapeutic emphases of medication and DBS. Subsequently, AI-based video analysis was applied to quantify kinematic parameters, including amplitude, frequency, and coefficients of variation, using AI algorithms (NERVTEX Co. Ltd.). Comparisons were made for levodopa effects (Levodopa-off vs. Levodopa-on), DBS effects (OFFMED/OFFSTIM vs. OFFMED/ONSTIM), and therapy-specific differences (Levodopa-on vs. OFFMED/ONSTIM). RESULTS: Conventional UPDRS-III item scores suggested that levodopa was more effective than DBS in improving upper-limb tasks (items 3.4 Finger Tapping and 3.5 Fist-clenching test), while lower-limb tasks (items 3.7 Toe Tapping and 3.8 Leg Agility) showed no significant changes. In contrast, AI-based kinematic analysis revealed more differentiated treatment effects. Levodopa was associated with improvements in movement speed, amplitude, and stability in the upper limbs, as well as a significant impact on lower-limb amplitude, both in toe tapping (item 3.7) and leg agility (item 3.8). DBS, by comparison, enhanced upper-limb motor output but had limited effects on the lower limbs, with improvements in speed and amplitude observed only in the toe tapping (item 3.7) task. Additionally, levodopa demonstrated superior improvements in lower-limb amplitude, both in toe tapping (item 3.7) and leg agility (item 3.8), compared to DBS. CONCLUSION: This study demonstrates that AI-based kinematic analysis enables a nuanced and individualized characterization of motor responses to medication and STN-DBS in Parkinson's disease, complementing conventional clinical scoring. Although both therapies improve bradykinesia, they appear to preferentially modulate distinct motor domains across individuals, underscoring their complementary roles in treatment. These findings highlight the potential of AI-based motor assessment to support personalized symptom profiling and more individualized therapeutic decision-making in Parkinson's disease.