| Abstract: | Owing to the complexities involved in obtaining direct measures of in vivo muscle forces, validation of predictive models of muscle activity has been difficult. An artificial neural network (ANN) model had been previously developed for the estimation of lumbar muscle activity during moderate levels of static exertion. The predictive ability of this model is evaluated in this study using several techniques, including comparison of response surfaces and composite statistical tests of values derived from model output, with multiple EMG experimental datasets. ANN-predicted activation levels were accurately modelled to within 3% across a range of experiments and levels of combined flexion/extension and lateroflexion loadings. The results indicate both a high degree of consistency in the averaged muscle activity measured in several different experiments, and substantiate the ability of the ANN model to predict generalized recruitment patterns. It also is suggested that the use of multiple comparison methods provides a better indication of model behaviour and prediction accuracy than a single evaluation criterion. |
