Objective:

To demonstrate the capabilities of video-based gait analysis to extract clinically meaningful kinematic features of neurotypical toddlers walking.

Background:

Laboratory-based pediatric gait analysis in neurology enables insights into underlying pathophysiology, guides surveillance, and informs treatment decision-making, but is not accessible to most U.S. children. We propose a video-based method of pediatric gait analysis drawing upon advances in computer vision to extract key kinematic features in a cohort of toddlers. Our long-term goal is to use video-based approaches to gait analysis to improve early identification, monitoring, and treatment planning of gross motor impairment in pediatric neurology.

Design/Methods:

We conducted a retrospective observational study of neurotypical toddlers walking on a straight path. We applied markerless pose estimation onto videos using OpenPose, an open-source machine learning model. We adapted an existing MATLAB pipeline to generate kinematic features such as step length, gait speed, and step time from key body features. Univariate linear regression analyses were conducted in R on trial-level gait parameters to evaluate for effects of age (in months) and sex.

Results:

We analyzed sagittal videos of 83 neurotypical toddlers (42 females, 41 males, 13 to 19 months old). Participants completed 5.72 trials on average. We saw a trend towards significance for older children ambulating with an increased step length (β=-0.257, p=0.068). Males had an increased step length (β =1.461, p=9.28×10−5) and velocity (β =0.08, p=6.75×10−4) compared to females.

Conclusions:

We show initial evidence that these kinematic gait features are readily measurable from video and capture subtle differences in gait based on age and biological sex. Our ongoing work will assess the validity of video-based gait features with synchronously collected data from a pressure-sensitive mat and prospectively evaluate videos of children with neurogenetic disorders compared to clinical scales.