In this thesis, we developed a stereo-visual hand-tracking system enabling 3D
reconstruction of the trajectories of 21 anatomical landmarks of the hand during
the Nine-Hole Peg Test (NHPT), a standardized tool for assessing fine motor
function of the upper limbs that in its traditional form records only the
total completion time, thereby overlooking information about movement
quality.
The system is based on two synchronized RGB cameras calibrated in a common
reference coordinate system defined by the test board, combined with the
MediaPipe Hands tool for landmark detection in individual frames. The detected
2D coordinates were reconstructed into 3D space using triangulation. Based on
the reconstructed trajectories, we computed a set of kinematic features
describing spatial, dynamic, and coordinative properties of hand movement, and
selected 24 key features for further analysis through dimensionality reduction.
Clinical evaluation was performed on a final cohort of 167 patients with multiple sclerosis, with the EDSS analysis including 162 patients and the CogEval analysis including 107 patients. Results of regression and
correlation analyses show that higher disability is most strongly associated with
prolonged task completion time and with increased linearity and reduced jerkiness
of thumb movement. For cognitive status, task execution speed emerged as the
primary predictor. The proposed approach is non-invasive and cost-effective,
preserves the standard NHPT protocol, and complements the traditional time
measure with additional information on movement quality.
|
