An efficient framework for fully deformable finite element musculoskeletal spine modeling: development, sensitivity analysis, and validation.

Abstract

Computational musculoskeletal models are fundamental tools for investigating spinal biomechanics. However, currently available approaches still rely on major simplifications. In particular, fully deformable finite element (FE) models capable of quantifying load-sharing across all spinal structures rarely incorporate the action of muscles. As a consequence, detailed predictions of internal loads deriving from realistic loading conditions still lack. This study integrates an active muscle architecture into a lumbar FE model using a multi-stage dimensionality reduction strategy, combining Plackett-Burman screening, Latin Hypercube, and Farthest Point Sampling, to develop surrogate Generalized Linear Models. These models, built on a reduced set of simulations, efficiently predict spinal kinematics as a function of muscle forces. Using these surrogates, an optimization procedure identified the muscle activations that satisfy kinematic targets while minimizing an energy criterion. The results accurately captured physiological trends: abdominal activations and intradiscal pressure increased with trunk flexion. Compressive forces (CFs) at L4-L5 in standing were 74 % of body weight, within the 58 %-98 % literature range. The model's full deformability also allowed for a detailed load-sharing assessment. For instance, in standing, the disc consistently carried over 44 % of CFs, followed by facet capsules (FCLs) contributing up to 26 %. During flexion, disc load nearly doubled, while FCLs became markedly tensioned (reaching 49 % at 20°). Also, shear forces (anterior 31 % to posterior 17 % of body weight) matched literature values, with the disc providing the primary contribution. These findings confirm the model's ability to predict complex loading patterns while reducing the computational cost of musculoskeletal FE simulations.