Accessible 2D video-based system for gait kinematic analysis: an inter-rater reliability study

Quantitative gait analysis supports clinical decision-making and outcome monitoring, yet its use remains largely restricted to specialized laboratories due to cost and operational complexity. Smartphone-based two-dimensional video approaches using open-source software have shown promising validity, but inter-rater reliability-particularly the influence of examiner expertise across spatiotemporal and kinematic outcomes-remains insufficiently characterized in powered samples.

In this observational study, 84 healthy adults were included and recorded during level barefoot walking using a standardized smartphone setup. Four examiners (two experienced, two non-experienced) independently extracted spatiotemporal parameters and hip, knee, and ankle joint angles at different gait events using Kinovea. Inter-rater reliability was assessed with intraclass correlation coefficients (two-way mixed-effects, absolute agreement; ICC [3,1]) with 95% confidence intervals, and measurement error was quantified using the standard error of measurement (SEM) and minimal detectable change (MDC95). Agreement between expert and non-expert examiners was explored using Bland-Altman plots for representative variables. Mean values were descriptively compared with established three-dimensional reference values.

Global spatiotemporal descriptors demonstrated excellent reliability with minimal error and limited dependence on examiner expertise (walking speed ICC = 0.96; cadence ICC = 0.98; stride length ICC = 0.96). In contrast, temporal gait phase variables showed poor reliability overall (ICC ≈ 0.31-0.43), with higher-but still limited-agreement among experienced raters. Joint kinematics exhibited joint- and event-dependent reliability following a proximal-to-distal gradient: hip angles showed the highest consistency (ICC up to 0.88), knee angles moderate-to-good agreement (ICC ≈ 0.65-0.85), and ankle angles the greatest variability, particularly for event-dependent measures (e.g., opposite toe-off ICC = 0.45). Bland-Altman analyses revealed negligible bias for spatiotemporal outcomes but wider limits of agreement for kinematic variables. Mean values were biomechanically coherent, although systematic differences in joint angles were observed when compared with 3D reference data.

In healthy young adults, a smartphone- and Kinovea-based workflow can provide reliable spatiotemporal gait metrics and moderate-to-good reliability for selected proximal kinematics, whereas temporal gait phase variables and distal joint angles remain less reliable and should be interpreted cautiously in longitudinal or multi-examiner contexts.