Shape model constrained scaling improves repeatability of gait data.

Decisions made by gait researchers in the generation of kinematic or musculoskeletal models are a potential source of variation between researchers, leading to variable model outcomes. Statistical shape models can accurately predict bone geometry and have the potential to improve the repeatability of clinical gait analysis. The purpose of this study was to determine if using a shape model to scale segment length and joint centre locations would improve repeatability of kinematic and kinetic gait data, compared to linear scaling methods. Five participants completed a motion capture experiment, including a standing static trial and walking at a self-selected speed. Anatomical landmarks from the static trial were used by five experienced researchers to generate kinematic models using two methods; (1) linear scaling in OpenSim, and (2) shape-model scaling using our 'MAP Client' scale tool. The resulting models were used to perform an inverse kinematic and inverse dynamic analysis on the walking trials, and variation between researchers was analysed by comparing outputs from the same motion capture trial using different models. Higher variability between researchers was observed in joint angles (P < 0.001), joint moments (P < 0.005), and joint powers (P < 0.005) when using linear scaling, compared to shape-model scaling. Variation was at least three times as large for linearly-scaled models compared to shape-model scaled models. We have identified that linear scaling can lead to substantial variability in gait data across researchers, even with the same experimental data. Using a shape model to scale musculoskeletal models results in repeatable kinematic and kinetic gait data.