Motion tracking with automated pose estimator can enhance ankle-foot-orthoses alignment

Ankle-foot orthoses are commonly prescribed to address motor impairments in individuals with neurological disorders. Proper alignment of Ankle-foot orthoses, commonly assessed by the ground reaction force in relation to the knee joint center and the shank-to-vertical angle, is crucial for their effectiveness but is time-consuming. This study validates automated pose estimator DeepLabCut on measuring these metrics using 2D videos with force vector overlay in individuals with neurological disorders wearing ankle-foot orthoses.

Thirty subjects with neurological disorders wearing ankle-foot orthoses participated. DeepLabCut's performance was compared to 3D gait analysis (Vicon). Subjects were randomly divided into three groups to train (20 subjects) and evaluate (10 subjects) DeepLabCut models. The number of subjects with untrackable points of interests in their videos was determined. The mean difference, root mean square error, intraclass correlation and repeatability coefficient between DeepLabCut and Vicon were computed for videos with trackable points of interest.

Only two subjects for ground reaction force distance to the knee and four subjects for the shank-to-vertical angle had untrackable points of interest. Excellent agreement between DeepLabCut and Vicon was found for the ground reaction force distance to the knee (mean difference 0.1 mm, root mean square error 6.7 mm, intraclass correlation 0.99, repeatability coefficient 6.7 mm) and shank-to-vertical angle (mean difference 0.7°, root mean square error 1.8°, intraclass correlation 0.99, repeatability coefficient 1.6°).

This study demonstrates that DeepLabCut can accurately and efficiently assess ankle-foot orthosis alignment parameters, paving the way for a simple and straightforward 2D video setup with a force-vector overlay for automated ankle-foot orthosis alignment.