gms | German Medical Science

Objectives: Muscle force estimation could enhance clinical movement analysis by giving insight into causes of impairments and informing targeted treatments. However, muscle force estimation is not typically used in this way due to difficulties in validation and selection of clinically adequate models. Estimated muscle forces are depend on its muscle activations, which, on the contrary, can be compared to observed muscle activation (elctromyography EMG). This study seeks, therefore, to examine the clinical utility of two different mathematical models which estimate muscle activation. Results are validated against surface EMG.

Methods: Ten healthy adults (5 male/5 female, 28±5 years old, 1.72±0.08m, 69±12kg) were recruited from amongst students and staff population. A ten-camera motion capture system (Nexus, Vicon, 100Hz) was used with four force plates mounted into the walkway (Kistler, 1000Hz) to capture joint angles and ground reaction forces (GRF) across a range of five gait speeds reflecting healthy and pathological gait. Parallel, surface EMG (Noraxon, 1000Hz) was captured on eight muscles of the leg (tibialis anterior, gastrocnemius lateralis and medialis, soleus, vastus lateralis or medialis, rectus femoris, semitendinosus). Activation of these muscles are estimated with two different mathematical modelling approaches, static optimisation (SO) and computed muscle control (CMC) in OpenSim (musculoskeletal model gait2392), using experimental joint angles and GRF. Agreement (linear correlation r) and deviation (mean absolute error MAE) in onset, offset and magnitude between models and EMG are evaluated.

Results and conclusion: A general agreement in speed dependence was visually shown between estimated and observed muscle activations. However, mean deviations (MAE) between modelling approaches and EMG were wide spread across speeds (SO vs EMG: 15-68%, CMC vs EMG: 13-69%). Slower speeds resulted in average in smaller deviations (MAE < 30%) than faster speeds. Linear correlation was in average high (r > 0.5) for 11/40 (CMC) and 6/40 (SO) conditions (muscles X speeds) compared to EMG, while shank muscles resulted in better correlation (15/40) than thigh muscles (2/40).

Modelling approaches do not yet show sufficient consistence in estimated with observed muscle activation to support immediate application to clinical muscle force modelling. This may be because assumptions underlying muscle activation estimations (e.g. muscles' origin and insertion, muscles' maximum isometric force) are not yet sufficiently individualizable. Application of such tools might be possible for slower walking speeds while taking caution that slow velocity might alter natural gait pattern. Future research needs to find timely and cost efficient ways to scale musculoskeletal models for a better individualisation to facilitate future clinical implementation. Surface EMG can be used as a validation tool subject to the condition that good signals are guaranteed and sensor placements are checked carefully in advance.