A neural network model for simulation of torso muscle coordination

An artificial neural network (ANN) was created to simulate lumbar muscle response to static moment loads. The network model was based on an abstract representation of a motor control system in which muscle activity is driven primarily to maintain moment equilibrium. The network model parameters were obtained by an iterative method (trained), using a modification of the standard backpropagation algorithm and moment equilibrium constraints. In contrast to previous ANN models of muscle activity, patterns of muscle activity are not target (training) values, but rather emerge as a result of moment equilibrium constraints. Assumptions regarding the moment generating capacity of muscles and competitive interactions between muscles were employed and enabled the prediction of realistic patterns of muscle activity upon comparison with experimental electromyographic (EMG) data sets (r2: 0.4-0.9). The success of the simulation model suggests that a motor recruitment plan can be mimicked with relatively simple systems and that ‘competition’ between responsive units (muscles) may be intrinsic to the learning process. Prediction of alternative recruitment patterns and differing magnitudes of co-contractile activity were achieved by varying competition parameters within and between units.