Design of personalized scoliosis braces based on differentiable biomechanics-Synthetic study

This work describes a computational methodology for the design of braces for adolescent idiopathic scoliosis. The proposed methodology relies on a personalized simulation model of the patient's trunk, and automatically searches for the brace geometry that optimizes the trade-off between clinical improvement and patient comfort. To do this, we introduce a formulation of differentiable biomechanics of the patient's trunk, the brace, and their interaction. We design a simulation model that is differentiable with respect to both the deformation state and the brace design parameters, and we show how this differentiable model is used for the efficient update of brace design parameters within a numerical optimization algorithm. To evaluate the proposed methodology, we have obtained trunk models with personalized geometry for five patients of adolescent idiopathic scoliosis, and we have designed Boston-type braces. In a simulation setting, the designed braces improve clinical metrics by 45% on average, under acceptable comfort conditions. In the future, the methodology can be extended beyond synthetic validation, and tested with physical braces on the actual patients.