LAUSR

Background Customized foot orthoses (FOs) are designed based on foot posture and function, while the interaction between these metrics and FO deformation remains unknown due to technical problems. Our aim was to predict FO deformation under dynamic loading using an artificial intelligence (AI) approach, and to report the deformation of two FOs of different stiffness during walking. Methods Each FO was fixed on a plate, and six triad reflective markers were fitted on its contour, and 55 markers on its plantar surface. Manual loadings with known magnitude and application point were applied to deform “sport” and “regular” (stiffer) FOs in all regions (training session). Then, 13 healthy male subjects walked with the same FOs inside shoes, where the triad markers were visible by means of shoe holes (walking session). The marker trajectories were recorded using optoelectronic system. A neural network was trained to find the dependency between the orientation of triads on FO contour and the position of markers on its plantar surface. After tuning hyperparameters and evaluating the performance of the model, marker positions on FOs surfaces were predicted during walking for each subject. Statistical parametric mapping was used to compare the pattern of deformation between two FOs. Results Overall, the model showed an average error of <0.6 mm for predicting the marker positions on both FOs. The training setup was appropriate to simulate the range of triads’ displacement and the peak loading on FOs during walking. Sport FO showed different pattern and significantly higher range of deformation during walking compared to regular FO. Conclusion Our technique enables an indirect and accurate estimation of FO surface deformation during walking. The AI model was capable to make a distinction between two FOs with different stiffness and between subjects. This innovative approach can help to optimally customize the FO design.