Authors (including presenting author) :
TSOI KM, YU JKP, CHAN BCC, HO AH, CHAN MKL, MA EWL
Affiliation :
Hospital Authority Community Rehabilitation Service Support Center (CRSSC)
Keyword 1: :
Rehabilitation
Keyword 2: :
Markerless Motion Analysis
Keyword 3: :
Computer Vision
Keyword 5: :
Joint Kinematics
Introduction :
Quantitative assessment of hand and finger kinematics is essential for diagnosis, treatment planning, and monitoring rehabilitation outcomes in patients with conditions such as tendon injuries, osteoarthritis, and post-stroke motor impairment. Conventional tools, including manual goniometry and observational scales, are labour-intensive, examiner-dependent, and limited in their ability to capture complex, dynamic three-dimensional (3D) motion of individual digits. These limitations contribute to reduced measurement precision, poor inter-rater agreement, and practical barriers to routine use in clinical environments.
Objectives :
There is therefore a clear need for an objective, affordable, and easy-to-use system capable of providing detailed, repeatable 3D kinematic data during functional hand movements.
Methodology :
The HandCube is a markerless system using a four normal webcam setup for multi-view to reconstruct 3D hand. During evaluation, a patient is required to perform nine different gestures within the cube (hand open, fist, oppositions and pinches). Based on computer vision and machine learning algorithm (Mediapipe-based hand landmark recognition advanced stereoscopic analysis), prominent joint landmarks in 3D space are automatically output as kinematic report, including Metacarpophalangeal (MCP), Proximal Interphalangeal (PIP), and Distal Interphalangeal (DIP) joint angles, as well as interdigital web space range. Before analysis, custom smoothing and rotation-aware transforms are applied so the coordinates are stable and consistent across views. These cleaned data are then converted into joint-angle measures. On top of that, downstream routines segment each repetition using marker timelines, compute range of motion (ROM) of joint angles in series of gestures, and generate annotated videos that overlay time-series joint angle changes on the original footage. A Python-based desktop app ties everything together: it guides users through data collection, launches batch analyses with a few clicks, and presents interactive result tables for clinicians to review. To verify the system's reliability, a calibrated robotic hand (OHand, OYMotion) was used in this study to perform the series of hand motion, the actual angles of joints (measured by goniometer) had been matched and compared to angles provided by hand cube system. The HandCube recorded each gesture five times while the robotic hand was rotated through 360 degrees to simulate different orientations.
Result & Outcome :
the HandCube system demonstrated strong agreement with researchers using a goniometer. The max absolute error for joint angles was 9.6°, which is less than 10°. The system also provides 0.951 in average Intraclass Correlation Coefficient (ICC) among all movable finger joints of the robotic hand, which showed excellent test-retest across repeated measurements. The HandCube represents a significant advancement in rehabilitative assessment technology. It provides a highly objective, easy to use, affordable and efficient method for quantifying hand kinematics, addressing the limitations of subjective manual methods. This innovation has the potential to standardize hand assessment, enable manageable progress tracking, and support data-driven clinical decision-making. Future work will focus on validating the system in larger cohorts and exploring its application in different clinical environment.
