To investigate human voluntary muscle contraction, computational neuromuscular models are commonly modelled with a single Hill-type muscle actuator receiving a unique experimental neural control from electromyography (EMG). Despite countless successful applications, this approach is limited as it cannot describe nor control the recruitment and neuromechanical dynamics of the individual motor units (MUs) constituting the muscle. As detailed in the first systematic review of the field reported in this thesis, current Hill-type modelling approaches are therefore limited to simulate and investigate neuromuscular control during human voluntary muscle contraction.
The aim of this thesis was to develop a novel neuromuscular model described as a collection of Hill-type MU actuators and controlled by adequate experimental neural inputs to allow the forward simulation of the MU pool dynamics during human voluntary muscle contraction.
A large portion of this thesis focused on the experimental and computational methods necessary to obtain reliable neural inputs to the modelled MUs. Motoneuron (MN) spike trains were experimentally identified from the decomposition of high-density EMG (HDEMG) signals. With refinements in the experimental and processing protocols, the number and quality of the identified spike trains were maximized. Despite that, only small portions of the active MU pool were identified, and the HDEMG spike trains did not provide an accurate approximation of the real neural command to muscle. To address this limitation, a novel computational pipeline was developed to infer from the HDEMG spike trains the discharge activity of the complete MN population. The pipeline used mathematical relationships between eight MN electrophysiological and anatomical properties, that were derived for the first time from the processing of an extensive set of mammal studies. The new extrapolated set of discharging MN proved to both accurately replicate the HDEMG data and accurately approximate the neural command to muscle.
The reconstructed MN spike trains were input to a state-of-the-art neuromuscular model to control its collection of Hill-type MUs. In doing so, this MN-driven model intrinsically accounted for the dynamics of MU recruitment and forward simulations of human voluntary muscle contraction were performed. The model was scaled with subject-specific musculoskeletal quantities and the predicted muscle force, obtained with advanced models of the MU neuromechanical properties, was experimentally validated.
By bridging the gap between recently available experimental data at the MN level and Hill-type neuromuscular modelling, this thesis provides a state-of-the-art tool for investigating neuromuscular control during human voluntary contractions, with possible important applications in neurorehabilitation and human-machine interfacing.