Empirical Evaluation of Models Used to Predict Torso Muscle Forces

Abstract

Risk factors associated with low back pain (LBP) have been identified by previous epidemiological research. Among these factors, the type and amount of loading required by a work task exhibit a direct relationship with the occurrence of LBP. Based on these findings, biomechanists have attempted to model the behavior of the lumbar region of the spine, with the objective of determining the distribution of loads on the different tissues of the region, especially in the muscles surrounding the lumbar spine. Knowledge of the loads in these tissues would allow for subsequent contrast against estimated tissue tolerance levels. The prediction of muscle forces is of particular importance given their large estimated magnitudes and contribution to spine loads. Two different approaches have historically been taken to predict muscle forces. The first relies on electromyography (EMG), while the second attempts to predict muscle responses using mathematical models. Three such predictive models are compared here: optimization-based Sum of Cubed Intensities, Artificial Neural Networks, and Distributed Moment Histogram. These three models were adapted to run using recently published descriptions of the lower back anatomy. To evaluate their effectiveness, model predictions were compared to a database of muscle activations obtained from 14 muscles of 8 participants (4 male and 4 female) with similar height and weight. The participants resisted loads applied to their torso via a harness. Eleven different 3-dimensional static loading conditions were applied, with associated magnitudes of 50% and 90% of each participant's maximum torso strengths. Results showed the models performed poorly (average R²'s on the order of 0.4), and that none of the models was clearly superior in terms of predictive ability. Interpretation of the models' output in their present state should be made cautiously when evaluating loading conditions that exceed 50% of an individual's torso strength. Discrepancies that were found between internal moments (at L3/L4) determined empirically and moments measured with a force plate suggest possible inaccuracies in model parameters including maximum muscle stress and/or anatomy.