Stroke is a leading cause of long-term disability, often affecting upper-limb motor function and requiring continuous assessment. The Fugl-Meyer Assessment (FMA), though a clinical gold standard, is time-consuming and demands specialized personnel. This study presents an AI-driven, low-cost rehabilitation exergame that simultaneously provides therapy and automatically estimates upper-limb motor performance during gameplay using only a standard camera. Sixteen kinematic and spatiotemporal features were extracted from 2D hand and arm trajectories of twelve post-stroke individuals (24 limbs, 14 affected) using the MediaPipe framework. Features such as hand angle, range of motion, movement area, traveled distance, and shoulder–elbow coordination showed strong correlations with FMA scores and stratified participants by motor severity. A lightweight linear regression model achieved high predictive performance (Spearman ρ = 0.92, R² = 0.89, RMSE = 4.42) and classified severity levels with 86–93% accuracy. This interpretable approach outperformed complex machine learning models, highlighting the clinical relevance of transparent metrics embedded in gameplay. The proposed framework is sensor-free, scalable, and reproducible, offering immediate feedback while reducing clinical workload and enabling accessible digital biomarkers for telerehabilitation and remote monitoring after stroke.

Stroke is a leading contributor to long-term disability worldwide, and rehabilitation often relies on costly devices, limited infrastructure, or labor-intensive protocols. While virtual reality-based exergames have gained popularity for promoting patient engagement, most rely on proprietary sensors or wearable electronics, limiting accessibility and clinical adaptability. This study presents the design, implementation, and pilot evaluation of a custom exergame that estimates the 3D elbow angle using a single RGB camera and two colored spheres as markers, eliminating the need for specialized hardware. The proposed system performs camera calibration, color segmentation, geometric 3D reconstruction, and real-time elbow angle estimation using low-cost equipment. Extensive technical tests revealed robust performance, with angular errors below 5° for joint amplitudes under 110°, and consistent accuracy across different lighting conditions, marker sizes, and distances. Additional tests showed that excessive sphere velocity (>20 cm/s) or proximity to image corners increased error due to motion blur and lens distortion, respectively. The system outperformed the AI-based MediaPipe framework in occluded-arm scenarios. Regression analysis showed strong correlation (r = 0.70) between movement velocity and angular error. Usability testing with eight post-stroke participants yielded a mean SUS score of 92.5/100. The proposed solution is a promising alternative for home-based, sensor-free rehabilitation, supporting personalized exercise routines and remote progress monitoring.

Background and Purpose: Accurate assessment of upper limb motor function is critical for guiding effective rehabilitation strategies in stroke survivors. The Fugl‐Meyer Assessment for Upper Extremity (FMA‐UE) is a widely used clinical tool, yet it is time‐consuming and subject to inter‐patient variability. A low‐cost, computer vision‐based exergame was developed to provide a remote rehabilitation alternative that also enables motor assessment. This approach represents a promising alternative for scalable and objective assessment, but requires validation. This study aims to evaluate the feasibility and preliminary validity of a computer vision‐based exergame system for assessing upper limb motor function in individual post‐stroke. Specifically, it investigates correlations between kinematic features derived from gameplay and FMA‐UE scores. It is expected that gameplay‐derived metrics, such as range of motion, jitter index, and average hand angle, will show moderate to strong correlations with the FMA‐UE, supporting the exergame's potential as a remote assessment tool. Methods: This is a cross‐sectional feasibility and preliminary concurrent validity pilot study with 15–30 individuals with chronic stroke. Participants will perform a markerless camera‐based exergame producing pseudo‐3D hand coordinates using MediaPipe‐based hand tracking. Kinematic features, including range of motion, average hand angle, and jitter index, will be extracted from gameplay. Each participant will also be evaluated using the FMA‐UE by a certified therapist. Correlation and regression analyses will assess the relationship between exergame metrics and clinical scores. Usability will be assessed using the System Usability Scale (SUS). Results: We expect moderate to strong correlations between selected gameplay metrics and FMA‐UE scores, particularly for range of motion, average hand angle, total traveled distance, jitter index, and score. The system is also expected to demonstrate high usability, indicating its feasibility for broader use in rehabilitation settings. Discussion: If validated, the proposed exergame may serve as an accessible and scalable digital tool to support future development of objective assessment tools in stroke rehabilitation, potentially reducing clinical workload and enabling remote evaluations. A future large‐scale study will evaluate test–retest reliability, longitudinal responsiveness, and home‐based implementation.

Background: Rigidity is a cardinal symptom of Parkinson’s Disease (PD), yet its clinical evaluation remains largely subjective and susceptible to errors. This study introduces an innovative method for objectively classifying individuals with PD by combining an active wrist orthosis with Machine Learning (ML) models. Methods: The orthosis, equipped with current and force sensors, recorded biomechanical signals during passive wrist flexion and extension, from which twelve quantitative features were extracted. Data were collected from 30 participants (15 with PD and 15 Healthy Controls). Nineteen supervised ML algorithms were systematically evaluated through feature selection, cross-validation, and hyperparameter tuning. Results: Using all twelve features, QDA achieved an accuracy of 0.889 and sensitivity of 1.000, followed by GPC (0.778) and LDA (0.778). After applying feature selection with the Correlation-based Feature Subset to reduce redundancy, Extra Trees reached 0.833 accuracy, while both QDA and GPC maintained accuracies of 0.778. This consistency across models, even with a reduced feature set, highlights the robustness of the extracted biomarkers. Conclusions: These findings confirm that wrist rigidity signals provide discriminative quantitative information between PD patients and HC and are able to support PD classification, combining engineering innovation with clinical practice that highlights the potential of integrating wearable devices and ML as a personalized healthcare in PD.