An integrated multi-variable optimization approach to tailor ankle-foot orthosis stiffness to end-user needs

Background: Ankle foot orthosis (AFOs) are devices commonly prescribed to assist or rehabilitate gait. A critical parameter influencing their effectiveness is the stiffness of the AFO. Although suppliers typically recommend stiffness levels based on general factors such as body weight and activity level, these guidelines are insufficient to achieve optimal stiffness tailored to each individual. In this work, we introduce an integrated multi-variable optimization approach that simultaneously considers multiple aspects of gait. Unlike previous approaches that rely on a single performance metric (e.g., metabolic cost) or impose a predefined hierarchy among gait parameters, our method makes no assumptions on relative gait domain importance. Furthermore, it allows the inclusion of users’ priorities, enabling a more personalized optimization of AFO stiffness. Methods: Ten children with cerebral palsy (CP) participated in an experimental protocol using the inGAIT-VSO AFO, completing five separate 2-minute walking trials, each with a different stiffness configuration. To determine the optimal stiffness for each participant, our method evaluated performance across five key gait domains: kinematics, spatio-temporal, balance, user perception, and muscular control. Furthermore, we investigated the integration of physiotherapists’ and users’ priorities into the optimization process, and explored the potential of a deep learning model to simplify future data collection needed for the optimization. Results: The proposed optimization method identified the stiffness configuration for each child with CP that most closely aligned their gait to healthy patterns considering the five gait domains. The optimal stiffness varied not only across participants but also across gait domains within the same participant. These findings reveal the importance of having a multi-variable, user-tailored approach. Overall, the inclusion of physiotherapists’ and users’ priorities did not alter the optimal stiffness selection. Conclusion: Our proposed optimization approach opens new possibilities for future research into the personalization and fine-tuning of AFO stiffness. It may also benefit from expanded data collection efforts that enable a more robust evaluation of the proposed deep learning model, supporting its integration into clinical practice.