A multimodal computer-vision based pipeline for eye-hand coordination quantification during the Nine-Hole Peg Test
The Nine-Hole Peg Test (9-HPT) is a widely used clinical measure of manual dexterity in multiple sclerosis (MS), but its primary outcome, task completion time, provides limited information about the visuomotor mechanisms contributing to impaired performance. Instrumented approaches can provide richer kinematic information but often require modifications to the task or physical instrumentation that may alter natural hand-object interaction. We developed a multimodal computer-vision pipeline that enables synchronized analysis of gaze, hand kinematics, and object state during an unmodified 9-HPT. Gaze was recorded using Pupil Labs Invisible glasses, three-dimensional hand landmarks were reconstructed using WiLoR, and a custom YOLO26x-seg model detected and tracked the pegboard, holes, pegs, and collection bin. Eye-tracking, hand, and object data were temporally synchronized at the video-frame level and spatially registered to a common pegboard-centered coordinate system using frame-specific projective transformations. The pipeline further integrated hole-state transitions, hand-zone classification, and fingertip kinematics to identify PICK and PLACE events and segment hand transport and stationary phases, while fixation and saccade labels were derived from the eye-tracking system. Automated detections were visually reviewed and corrected when necessary. The resulting framework provides synchronized, event-resolved measures of gaze behavior, hand movement, object interaction, and eye-hand coordination while preserving standardized 9-HPT administration. This approach extends the 9-HPT beyond completion time and provides a reproducible platform for investigating the visuomotor mechanisms underlying dexterity impairment in MS and other neurological disorders.