Performance of an Artificial Neural Network Model in the Prediction of Lower Torso Muscle Recruitment Patterns

Abstract

The prediction of muscle forces around the low back is a key aspect in the calculation of spinal loads. Artificial Neural Networks (ANNs) are suited to handle this prediction problem, and have been employed in the past in similar efforts. This paper evaluates the performance of an ANN in predicting trunk muscle forces, whose values were estimated from measurements of myoelectric activity during various high-magnitude static trunk exertions. Sensitivity of the ANN to its two user-adjustable parameters, Muscle Self-Inhibition and Muscle Inhibition, is also evaluated. Results indicate moderate agreement between predicted forces and forces estimated from measured myoelectric activity (average R²s in the 0.50s), although some muscles were well predicted (R²s in the 0.80s). The model iterations resulting from variations in the ANN Inhibition parameters varied only slightly in terms of prediction performance levels, suggesting that this particular ANN model may be limited by other factors, including inter-individual differences in motor control strategies.