Modeling muscle function using experimentally determined subject-specific muscle properties

Muscle models are commonly based on intrinsic properties pooled across a number of individuals, often from a different species, and rarely validated against directly measured muscle forces. Here we use a rich data set of rat medial gastrocnemius muscle forces recorded during in-situ and in-vivo isometric, isotonic, and cyclic contractions to test the accuracy of forces predicted using Hill-type muscle models. We identified force-length and force-velocity parameters for each individual, and used either these subjectspecific intrinsic properties, or population-averaged properties within the models. The modeled forces for cyclic in-vivo and in-situ contractions matched with measured muscle-tendon forces with r2 between 0.70 and 0.86, and root-mean square errors (RMSE) of 0.10 to 0.13 (values normalized to the maximum isometric force). The modeled forces were least accurate at the highest movement and cycle frequencies and did not show an improvement in r2 when subject-specific intrinsic properties were used; however, there was a reduction in the RMSE with fewer predictions having higher errors. We additionally recorded and tested muscle models specific to proximal and distal regions of the muscle and compared them to measures and models from the whole muscle belly: there was no improvement in model performance when using data from specific anatomical regions. These results show that Hill-type muscle models can yield very good performance for cyclic contractions typical of locomotion, with small reductions in errors when subject-specific intrinsic properties are used.