Effect assessment of lifted orthopaedic shoes on gait in people with leg length discrepancy using a deep learning gait detection in the customisation process

Background Gait deviation by leg length discrepancy limits development of motor skills and causes lower limb injuries and pains. Orthopaedic shoes (OSs) have been a widely used treatment for the gait problems. However, their effects on gait performance at their customisation have not been widely studied due to the high complexity and cost in measuring and analysing gait changes. It misses an opportunity for providing optimally lifted OSs to individuals. This study therefore aimed to assess the effects of OSs through simple gait pattern analysis using a vision-based deep learning approach and provide a useful guideline for their customisation. Methods Sixteen participants, having the left leg short, underwent walking on straight paths with and without their trial OSs, initially lifted for equalising bilateral leg lengths. The vision-based deep learning model was employed to extract spatiotemporal gait parameters from the participant’s gait videos. Using the parameters, we examined pattern changes between the left and right gaits in terms of harmony, symmetry, regularity, and stability defined in this study. The gait pattern changes were evaluated using paired t -tests. Results With the trial OSs, significant improvement ( p <0.05) of the gait harmony was shown in the left gait. Conflicting pattern changes between the left and right gaits were observed in the gait symmetry and regularity analyses. The gait symmetry was significantly increased for step length ( p <0.05) whereas decreased for step phase ( p <0.05) with high variation and considerable gaps in the changes. The left gait became more regular with the increase in step length ( p <0.01) and phase on the contrary to the right gait. Regarding step phase, the overall gait regularity was significantly decreased ( p <0.05). The gait stability also showed a decreasing tendency. The overall gait performance with the trial OSs was counted as suboptimal, in which further individually-differentiated correction is required in their customisation. Conclusions This study raised additional considerations of examining individual gait performance when customising OSs and provided an avenue to develop evidence-based customisation strategies. The gait pattern analysis using a vision-based deep learning approach can be suggested as a feasible method for effective customisation of optimally corrected OSsfor gait rehabilitation.