Validation of a Markerless Non-Contact Gait Analysis System for Three-Dimensional Gait Kinematics in Patients with...

(1) Background: This study validated the accuracy of a markerless gait analysis system, Foot4D, against the Vicon 3D motion capture system in 60 patients with ankle injuries. (2) Methods: Synchronized gait data were collected using both systems while participants walked on a treadmill at self-selected speeds. The Foot4D system employs a Vision Transformer (ViT-H/16) encoder and the SKEL parametric model for markerless 3D pose reconstruction from multi-view depth cameras. Spatiotemporal gait parameters and lower limb 3D kinematics (hip, knee, and ankle) were compared using ICC, Bland-Altman analysis, and MAE. (3) Results: All 12 spatiotemporal parameters showed good-to-excellent agreement (ICC = 0.738-0.999, MAE = 0.009-0.108 m/s). Agreement for joint kinematics decreased from proximal to distal: hip (mean ICC ≈ 0.947, MAE 1.16-1.52°), knee (mean ICC ≈ 0.921, MAE 1.17-1.75°), and ankle (mean ICC ≈ 0.839). Ankle sagittal plane ROM demonstrated good agreement (ICC = 0.857-0.911), while the frontal (ICC ≈ 0.807) and transverse (ICC ≈ 0.816-0.835) planes were at the lower bound of good agreement, with a -3.19° systematic bias in the injured ankle sagittal plane. (4) Conclusions: Foot4D is clinically reliable for spatiotemporal and proximal joint kinematic assessment, though ankle multiplanar motion measurement warrants further algorithmic optimization.