Effects of Spinal Coupling and Marker Set on Tracking of Spine Models During Running

Abstract

Running is a complex motion producing many muscle and joint forces that cannot be directly measured. Musculoskeletal simulation enables estimation of these muscle and joint forces, but to have confidence these simulations and processes must be biofidelic. Most models include a rigid spine, but prescribing motion with a coupled spine model may allow more accurate inverse kinematics tracking of experimental data and allow truer resulting intervertebral force and muscle activation estimations. PURPOSE: To determine the effects of spinal coupling and the quantity of experimental data available on the tracking of experimental running data through use of a rigid and a coupled lumbar spine model. METHODS: Two subjects ran on a treadmill and 3 motion capture trials at different speeds were collected with 13 markers placed on the spine and 46 other markers placed over the rest of the body. The Full Body Lumbar Spine model has 30 degrees of freedom (DoF), including a lumbar spine with coupling constraints resulting in a net of 3 DoF among those 5 vertebral bodies. Two iterations of this model were used, one with the coupling of the lumbar spine enabled (CS), and one where the coupling was locked resulting in a rigid lumbar spine (RS). Inverse Kinematics (IK) was executed using six different combinations of spinal markers as tracking inputs for both models. The marker error after IK was computed at each frame, and the root-mean-square (RMS) error computed for the full trial. Effects of the model, subset of tracking markers used as input, and subject were analyzed with multiple regression and differences between tracking subsets were analyzed with Tukey pairwise comparisons. RESULTS: Choice of model (CS or RS) had a significant effect on the RMS error of the markers (p<0.001). The average RMS error across all spinal markers was 1.35 ± 0.30 cm for the CS vs. 1.64 ± 0.29 cm for the RS. The multiple regression showed a significant effect of tracking subsets, and subject (p<0.001). Tukey pairwise comparisons showed that the two best tracking subsets were one weighting all 13 spinal markers and one weighting two lumbar markers (L2, L4), two thoracic markers (T10, T4), and the C7. CONCLUSION: The CS model exhibits lower RMS errors than the RS model, and this error can be further minimized by the inclusion of additional thoracic and lumbar spine markers.